OPINION OF THE COURT
BECKER, Circuit Judge.Defendants are two brothers, Antonio and Policai Collado, who lived in New York and supplied heroin to an extensive wholesale distribution organization based in Philadelphia. The government charged the Collados, other New York suppliers, and various members of the Philadelphia organization with conspiracy to distribute heroin in violation of 21 U.S.C. § 846 (1988). The Collados were- tried by jury in the district court for the Eastern District of Pennsylvania. The jury convicted them of the conspiracy count and of six counts of using the telephone to facilitate the sale of heroin in violation of 21 U.S.C. § 848(b) (1988).
Their consolidated appeals raise two challenges to the district court’s calculation of their base offense levels under the Sentencing Guidelines. First, they contend that the district court erred when it attributed to them quantities of heroin distributed by their co-conspirators. This contention requires us to explicate and apply the relevant conduct provision of the Sentencing Guidelines, U.S.S.G. § 1B1.3, to three separate types of attribution: (1) attribution to both brothers of amounts distributed by the conspiracy before they joined it; (2) attribution to one brother of amounts the other brother supplied to the conspiracy; and (3) attribution to both brothers of amounts supplied to the conspiracy by other conspirators. We conclude that the district court properly attributed to one brother amounts distributed by the other. On the basis of this record, we cannot determine whether the first and third types of attribution were appropriate, however. Accordingly, we will vacate the judgment and remand for further development of the record and for resentencing.
The Collados also challenge the sufficiency of the evidence regarding the quantities involved in two specific transactions. We hold that the evidence regarding one of these transactions was insufficient to support the probation officer’s estimate, which was accepted by the district court, of the amount of heroin involved. Accordingly, we will direct the district court to exclude that amount from its calculations on resen-tencing.
I. FACTUAL BACKGROUND
A. The Conspiracy
The evidence presented at trial showed that the Collados were part of a wholesale heroin distribution organization that operated from two grocery stores located in North Philadelphia.. The Philadelphia conspirators obtained heroin from New York suppliers, including the Collados, and then sold it to others who intended to resell it.
On April 12, 1988, an agent of the Philadelphia DEA, Detective Fred Chavez, made the first of a series of undercover heroin purchases at one of the grocery stores. Chavez made three more purchases from the Philadelphia dealers over the next several months. The amounts purchased in the four transactions totalled 40.873 grams.
A court-authorized Title III interception of phone calls to and from one of the grocery stores was initiated on September 21, 1988. The surveillance continued until October 20, 1988, when the phone at the store was disconnected for eight days. The interception resumed on October 28, 1988, and continued until November 9, 1988.
*989It was through this telephone surveillance that the detectives identified Antonio and Policai Collado as New York suppliers of the Philadelphia dealers. Between September 21, 1988 and November 5, 1988, the Collados were in frequent contact with Pedro Eliado Frias, also known as Laito, an alleged manager of the Philadelphia organization who oversaw operations at one of the two grocery stores.1
On October 13, 1988, Detective Chavez and another undercover agent travelled to New York with two members of the Philadelphia organization. There they purchased 125 grams of heroin from Buena Ventura Marte. On October 15, 1988, Philadelphia police arrested Ramon Rivera, who allegedly managed operations at the second grocery store.2 Sixteen grams of heroin were seized from Rivera upon his arrest. The Collados were arrested in New York on August 1, 1989. No drugs were seized from them or found on the premises where they were arrested.
To tie the Collados to the conspiracy, the government primarily relied on two types of evidence: (1) the testimony of Rosa German, a member of the conspiracy turned government informant, and (2) tape recordings of the intercepted calls. German, who lived with Frias at the time of the alleged conspiratorial acts, testified that the Colla-dos had both supplied the Philadelphia dealers with heroin. The government’s expert witness testified that the conversations between the Collados and Frias reflected drug transactions in which the Collados agreed to supply Frias with heroin. Although the indictment charged the Collados with participating in the conspiracy beginning in April 1988, the government produced no direct evidence that the Collados were involved before late September 1988.
B. The Sentencing
As we explain in detail at pages 990-91 [typescript at 9-10], the base offense level for a defendant convicted of conspiring to distribute heroin depends upon the amount of drugs deemed “relevant” to the offense. In determining the Collados’ base offense levels, the probation officer considered the following amounts to be relevant to the offense: the quantity police purchased from the organization in Philadelphia (40.-873 grams); the amount seized from conspirator Ramon Rivera upon his arrest (16 grams); the amount purchased from Buena Ventura Marte (125.4 grams); and an estimate of the quantity of heroin discussed by the conspirators in intercepted phone conversations (686.5 grams). The probation officer concluded that from April to November, 1988, the conspiracy distributed 868.773 grams of heroin. Using this total, the probation officer calculated the base offense level for both Collados as 30.3
At sentencing, the Collados argued that the presentence investigation report incorrectly attributed to them quantities of drugs dealt by their co-conspirators (1) before they joined the conspiracy and (2) after they joined the conspiracy, but of which they had no knowledge. More specifically, they contended that, although the government’s evidence indicated that they had supplied heroin to Frias on several occasions between September 21 and November 5, 1988, there was no evidence that either brother had been a member of the conspiracy before September 21, 1988, or that either had knowledge of or involvement with heroin that Frias purchased from others (such as Marte) after they joined the conspiracy.
The Collados also challenged the government’s estimate of the quantities involved in two specific transactions. The government estimated that the first challenged transaction involved 62.5 grams of heroin and that the second involved 125 grams. Defendants claimed that the government *990failed to present sufficient evidence to justify these estimates.
The district court accepted the probation officer’s calculation, concluding, without further explanation, that “the presentence report properly allocated the amount of drugs involved in this case.” Based upon an offense level of 30, the court sentenced each defendant to a term of incarceration of 105 months on the conspiracy count and 48 months on each of the six substantive counts, all sentences to run concurrently.
II. STANDARD OF REVIEW
The government contends that the district court’s calculation of the defendants’ base offense levels is a factual finding subject to the clearly erroneous standard of review. “When reviewing the sentencing decisions of the district courts, ‘[w]e exercise plenary review over legal questions about the meaning of the sentencing guidelines, but apply the deferential clearly erroneous standard to factual determinations underlying their application.’ ” United States v. Fuentes, 954 F.2d 151, 152-53 (3d Cir.), cert. denied, — U.S.-, 112 S.Ct. 2950, 119 L.Ed.2d 573 (1992), quoting United States v. Inigo, 925 F.2d 641, 658 (3d Cir.1991). Accordingly, we exercise plenary review over the legal standard for determining when the calculation of a defendant’s base offense level may include amounts of drugs distributed by persons other than the defendant.
Our review of the sufficiency of the evidence, an issue which arises both in regard to our application of the legal standard for calculating the base offense level and in regard to the amounts involved in the two challenged transactions, is governed by the clearly erroneous standard. Where, as here, the district court makes no independent findings of fact in relation to sentencing issues, but instead adopts the reasons set forth by the probation officer in the presentence investigation report, we view the report as containing the only findings of fact that support the court’s sentencing decision. United States v. Belletiere, 971 F.2d 961, 968-969 (3d Cir.1992).
III. GENERAL PRINCIPLES FOR DETERMINING RELEVANT CONDUCT IN CALCULATING THE DEFENDANT’S BASE OFFENSE LEVEL
A. Overview
In most narcotics cases to which the Sentencing Guidelines apply, the defendant's sentence depends to a great extent upon the quantity of drugs deemed “relevant” to the offense.4 See U.S.S.G. § 1B1.3. After the sentencing court determines the relevant amount of drugs, it refers to the Drug Quantity Table, U.S.S.G. § 2Dl.l(c), and chooses the base offense level that corresponds to that amount. The base offense level serves as a starting point; after making the initial determination regarding the base offense level, the sentencing judge adjusts the base offense level upward or downward to reflect other factors, such as acceptance of responsibility or role in the offense, and then selects a sentence that falls within the applicable guideline range for that level.
The relevant conduct provision of the Guidelines, U.S.S.G. § 1B1.3, directs the sentencing court to consider all conduct deemed relevant to the offense of conviction in determining the base offense level. Specifically, the provision directs the sentencing court to include in the base offense level certain amounts in addition to those amounts the defendant was convicted of distributing. At issue here is the portion of the relevant conduct provision that authorizes what has been referred to as "accomplice attribution,” see William W. Wilkins, Jr. & John R. Steer, Relevant Conduct: The Cornerstone of The Federal Sentencing Guidelines, 41 S.C.L.Rev. 495, 506 (1990). In this context, accomplice attribution is the inclusion of amounts of *991drugs possessed, distributed, manufactured, sold, or otherwise “handled” by persons other than the defendant in the calculation of the defendant’s base offense level.5
B. The Standard for Accomplice Attribution
■ Subsection (a)(1) of the relevant conduct provision sets forth the conditions for accomplice attribution. It provides, in pertinent part, that the defendant’s base offense level
shall be determined on the basis of ... all acts and omissions ... for which the defendant would be otherwise accountable. ...
U.S.S.G. § lB1.3(a)(l) (1988).
An earlier version of the application note accompanying this section provided:
If the conviction is for conspiracy, it includes conduct in furtherance of the conspiracy that was reasonably foreseeable by the defendant.
U.S.S.G. § 1B1.3, application note 1 (November 1, 1987 version). Effective November 1, 1989, the Sentencing Commission amended this application note to clarify the definition of conduct for which the defendant is “otherwise accountable.” See United States Sentencing Commission, Guidelines Manual Appendix C at 31. The amendment replaced the application note quoted above with the following:
In the case of criminal activity undertaken in concert with others ... the conduct for which the defendant “would be otherwise accountable” also includes conduct of others in furtherance of the execution of the jointly-undertaken criminal activity that was reasonably foreseeable by the defendant. Because a count may be broadly worded and include the conduct of many participants over a substantial period of time, the scope of the jointly-undertaken criminal activity, and hence relevant conduct, is not necessarily the same for every participant. Where it is established that the conduct was neither within the scope of the defendant’s agreement, nor was reasonably foreseeable in connection with the criminal activity the defendant agreed to jointly undertake, such conduct is not involved in establishing the defendant’s offense level under this guideline.
U.S.S.G. § 1B1.3, application note 1.
Some cases decided before the 1989 amendment interpreted the relevant conduct provision very broadly. See Phylis Skloot Bamberger & David J. Gottlieb, eds, Practice Under the New Federal Sentencing Guidelines .§ [B][2] at 20 (Prentice Hall, Supp.1992) (citing cases). In those cases, courts evidently decided whether accomplice attribution was appropriate by asking whether the amounts distributed by the co-conspirator (1) were distributed in furtherance of the conspiracy as described in the count of conviction and (2) were reasonably foreseeable by the defendant in that the defendant was or reasonably should have been aware that co-conspirators were likely to distribute such amounts.
The amended application note makes clear that the standard for accomplice attribution is significantly more stringent, however. The language of this note indicates that, rather than evaluating accomplice attribution in light of the scope of the conspiracy as described in the count of conviction and the defendant’s awareness of the possibility that co-conspirators would distribute amounts in addition to those amounts distributed by the defendant, courts should look to the defendant’s role in the conspiracy. Specifically, the note instructs courts to assess accomplice attribution by determining whether the co-conspirator’s conduct was “in furtherance of the ... jointly-undertaken ... activity” (as opposed to the conspiracy as described in the count of conviction), “within the scope of the defendant’s agreement,” and “reasonably foreseeable in connection with the criminal activity the defendant agreed *992to undertake.” U.S.S.G. § 1B1.3, application note 1 (emphasis added). In our view, this language indicates that whether a particular defendant may be held accountable for amounts of drugs involved in transactions conducted by a co-conspirator depends upon the degree of the defendant’s involvement in the conspiracy and, of course, reasonable foreseeability with respect to the conduct of others within the conspiracy.
Our interpretation of the relevant conduct provision is informed by a consideration of its purpose. As Judge Breyer has explained, the relevant conduct provision represents an important exception to the “charge offense” approach generally taken by the Guidelines. United States v. Wood, 924 F.2d 399, 403 (1st Cir.1992). Under a charge offense system, the sentence is based on the conduct constituting the offense for which the defendant was convicted. As the application note recognizes, in conspiracy cases the count describing the offense “may be broadly worded and include the conduct of many participants over a substantial period of time.” If all defendants were sentenced on the basis of the conduct charged in the offense of conviction, then the conspirator who directed and controlled the operation would be treated the same as a lesser participant.
In contrast, the amended application note states that “relevant conduct[ ] is not necessarily the same for every participant" in the conspiracy, thus indicating that in some instances, sentencing courts must distinguish between co-conspirators for the purposes of accomplice attribution. See Wilkins & Steer, 41 S.C.L.Rev. at 510 (observing that defendant in narcotics conspiracy may be held accountable for different quantity of drugs than his or her co-conspirator). The relevant conduct provision therefore recognizes that, while it is appropriate to hold a defendant who exhibits a substantial degree of involvement in the conspiracy accountable for reasonably foreseeable acts committed by a eo-conspirator, the same cannot be said for a defendant whose involvement was much more limited.
An example chosen by the Sentencing Commission to illustrate the meaning of the relevant conduct provision confirms our interpretation of the standard for accomplice attribution. That example states:
Defendants H and I engaged in an ongoing marihuana importation conspiracy in which Defendant J was hired to help offload a single shipment. Defendants H, I, and J are included in a single count charging conspiracy to import marihuana. For the purpose of determining the offense level under this guideline, Defendant J is accountable for the entire single shipment of marihuana he conspired to help import and any acts or omissions in furtherance of the importation that were reasonably foreseeable. He is not accountable for prior or subsequent shipments of marihuana imported by Defendants H and I if those acts were beyond the scope of, and not reasonably foreseeable in connection with, the criminal activity he agreed to jointly undertake with Defendants H and I (i.e., the importation of the single shipment of marihuana).
U.S.S.G. § 1B1.3, illustration e. Although the “prior or subsequent shipments” are clearly in furtherance of the larger conspiracy described in the count of conviction and Defendant J was or reasonably should have been aware that persons who import one shipment of drugs are likely to import others, the example indicates that those shipments are not attributable to Defendant J if they are “beyond the scope of the criminal activity he agreed to jointly undertake.” Thus, this example confirms our view that the crucial factor in accomplice attribution is the extent of the defendant’s involvement in the conspiracy.
C. Case Law
As we have already noted, the earlier cases interpreted the relevant conduct provision very broadly. Indeed, some of these cases easily could be read as holding that the defendant could be sentenced based on the entire amount of drugs involved in the conspiracy merely because the defendant had been found guilty of the crime of conspiracy. See, for example, United States v. Holland, 884 F.2d 354, 358 (8th Cir.*9931989). By contrast, the more recently decided cases reflect an approach consistent with our focus on the defendant’s role in the conspiracy.6
United States v. Edwards, 945 F.2d 1387 (7th Cir.1991), presents perhaps the most thorough and searching analysis of accomplice attribution to date. Although the Seventh Circuit defined the standard for accomplice attribution in terms of the factors described in the earlier version of the application note — furtherance of the conspiracy and reasonable foreseeability, id. at 1392,7 — it repeatedly acknowledged the importance of evaluating the defendant’s “level” or “degree” of “commitment” to the conspiracy. See id. at 1392, 1394. Moreover, in determining whether the amounts distributed by the conspiracy were properly attributed to each defendant, the court looked to the degree of the defendant’s involvement in the conspiracy. See, for example, id. at 1401 (amounts distributed by co-conspirators not attributable to defendant whose role in the conspiracy was “more limited”). Finally, Edwards reviewed the cases in which defendants have been held accountable for the entire amount of drugs distributed in the conspiracy and explained that in most of these cases, the facts indicated that the defendants were substantially involved in the conspiracy. Id. at 1394-95.
United States v. Carmody, 753 F.Supp. 917 (M.D.Ala.1990), aff’d without opinion, 940 F.2d 673 (11th Cir.1991), also comports with our interpretation of the standard for accomplice attribution. The three defendants in Carmody purchased drugs from a supplier who operated a large-scale manufacturing and distribution operation. The government contended that the defendants’ sentences should reflect quantities their supplier sold to others. Such quantities were “reasonably foreseeable” by the defendants, the government argued, because the defendants knew that their supplier produced and sold large quantities of drugs. The court rejected the government’s definition of reasonable foreseeability, reasoning that “if this were true, all the drugs sold by a dealer could be attributed to any and all of his customers, for a person who purchases narcotics should reasonably foresee that the dealer is likely also selling to others.” 753 F.Supp. at 920.
By refusing to define the reasonable foreseeability required for accomplice attribution as involving nothing more than the defendant’s awareness of the possibility that co-conspirators were distributing drugs to others, Carmody recognized the importance of evaluating the defendant’s role in the conspiracy. The court correctly observed that before conduct of one con*994spirator may be attributed to another at sentencing, that conduct must be “reasonably foreseeable ‘in connection with the criminal activity the defendant agreed to jointly undertake.’ ” Id., quoting U.S.S.G. § 1B1.3, application note 1 (emphasis in opinion). Moreover, to determine the scope of the criminal activity the defendants agreed to undertake, the court looked to the facts surrounding their involvement in the conspiracy. Although the court acknowledged that the defendants were aware of their supplier’s sales to others, it observed that none of them ever assisted or profited in those sales. Id., 753 F.Supp. at 921. Accordingly, the court refused to attribute the amounts involved in those sales to the defendants.
The results in United States v. North, 900 F.2d 131 (8th Cir.1990), which refused to allow accomplice attribution, and United States v. Garcia, 954 F.2d 12 (1st Cir.1992), which did permit it, are also consistent with our interpretation of the relevant conduct provision. North purchased drugs from Murphy to distribute to others. In addition, North was aware that Murphy also sold drugs to other dealers. The district court attributed to North the amounts Murphy attempted to sell to a confidential informant. The government argued that the attribution was proper because North and Murphy had conspired to distribute drugs to others; Murphy’s attempted sale to the confidential informant, the government contended, was part of that conspiracy.
The Eighth Circuit reversed because the evidence indicated that the conspiracy between North and Murphy “had a more limited objective, specific to Murphy and North supplying drugs to each other.” 900 F.2d at 133. Although North admitted that he knew Murphy sold drugs to others, the court observed that the amount Murphy attempted to sell to the confidential informant was obtained without North’s aid, “[t]he sale was arranged and executed without any knowledge or assistance on the part of North,” and "its success was not advantageous to North.” Id., at 134. The court also stated that there was no connection between the attempted sale and the transactions between North and Murphy. Accordingly, the court concluded that the attempted sale was “merely another part of Murphy’s distribution practice,” id., and therefore could not be attributed to North.
In our view, North was correctly decided because even though North was aware of Murphy’s dealings with others, those dealings were outside the scope of the activity North agreed to undertake with Murphy. Because North’s involvement in Murphy’s distribution operation was limited to the transactions between the two, amounts involved in other transactions should not have been included in the calculation of North’s sentence. See also United States v. Rivera, 898 F.2d 442, 445-46 (5th Cir.1990) (refusing to attribute to defendant amounts sold by others absent finding that defendant was part of jointly-undertaken scheme to distribute those amounts).
In Garcia, an undercover agent, Vega, told Garcia that he was looking for cocaine. Garcia introduced Vega to Zuleta, who sold Vega cocaine on several occasions. Vega and Zuleta met at the bar Garcia owned before each of the first three sales. After the first sale, Vega asked about obtaining a larger quantity at a cheaper price; Zuleta said that he would check with Garcia. After the second sale, Vega discovered that he had underpaid for the drugs. To remedy the problem, he found Garcia at the bar, explained the situation to him, and gave him the money he owed. When Vega arrived at the bar for the third sale, Garcia greeted him and indicated that he knew that Vega was there to meet Zuleta.
Zuleta made three more sales to Vega. None of these sales began at Garcia’s bar or appeared to involve him. Nevertheless, the district court found that the three later sales were part of a joint undertaking between Garcia and Zuleta and, accordingly, included the amounts involved in those sales in its calculation of Garcia’s base offense level.
The First Circuit upheld the district court’s decision. The court stated that “the measure of a defendant’s accountability for drug transactions in which he was *995not personally involved is usually congruent with the scope of his agreement with the other participants in the criminal enterprise.” 954 F.2d at 16. It observed that Garcia and Zuleta conspired to supply Vega with cocaine, that Garcia introduced Vega to Zuleta for that purpose, and that Garcia set the price. Accordingly, the court concluded that “Garcia’s commitment to the venture, and the scope of his conspiratorial agreement, encompassed the final series of transactions.” Id., at 17.
In our view, Garcia illustrates the kind of circumstances that justify accomplice attribution. Zuleta’s additional sales to Vega were reasonably foreseeable by Garcia in that Garcia knew or should have known that Zuleta was likely to continue supplying Vega. More importantly, those sales were within the scope and in furtherance of Garcia’s agreement with Zuleta, which had supplying Vega as at least one of its objectives.
In sum, the courts in Edwards, Carmo-dy, North, and Garcia all looked to the defendant’s role in the conspiracy in deciding whether to permit accomplice attribution. Although each case states the standard for accomplice attribution in a slightly different way, all are essentially consistent with our interpretation of that standard.8 In each case, the court permitted accomplice attribution only when the facts indicated that drug transactions conducted by the defendant’s co-conspirators were within the scope and in furtherance of the activity the defendant agreed to undertake.
D. Conclusion
As our description of the mechanics of sentencing indicates, see pages 991-992, the quantity of drugs attributed to the defendant usually will be the single most important determinant of his or her sentence. Accomplice attribution often results in a dramatic increase in the amount of drugs for which the defendant is held accountable, which translates directly into a dramatic increase in the sentence. As we have explained, whether an individual defendant may be held accountable for amounts of drugs involved in reasonably foreseeable transactions conducted by co-conspirators depends upon the degree-of the defendant’s involvement in the conspiracy. In assessing the defendant’s involvement, courts must consider whether the amounts distributed by the defendant’s co-conspirators were distributed “in furtherance of the ... jointly-undertaken ... activity,” were “within the scope of the defendant’s agreement,” and were “reasonably foreseeable in connection with the criminal activity the defendant agreed to undertake.” U.S.S.G. § 1B1.3, application note 1. We wish to emphasize that in deciding whether accomplice attribution is appropriate, it is not enough to merely determine that the defendant’s criminal activity was substantial. Rather, a searching and individualized inquiry into the circumstances surrounding each defendant’s involvement in the conspiracy is critical to ensure that the defendant’s sentence accurately reflects his or her role.
IV. APPLICATION OP THE STANDARD TO THIS CASE
The district court’s calculation of the Col-lados’ base offense levels reflects three separate types of accomplice attribution: (1) attribution to both brothers of amounts distributed by the conspiracy before they joined it; (2) attribution to each brother of amounts the other brother supplied to Fri-as; and (3) attribution to both brothers of the amounts supplied to the conspiracy by third parties.
A. Attribution of Amounts Distributed by the Conspiracy Before the Collados Joined It
The Collados contend that the amounts that Detective Chavez purchased from the Philadelphia dealers should not have been attributed to them because those *996purchases were made before they joined the conspiracy. Similarly, Antonio Collado argues that the amount involved in one of the sales Policai Collado made to Frias should not be attributed to him because there is no evidence that he was a member of the conspiracy when that sale occurred.
Detective Chavez made the first and second of the Philadelphia purchases in April 1988, the third in May 1988, and the fourth on September 13, 1988. Policai Collado made the sale that Antonio Collado claims should not be attributed to him sometime around September 21, 1988.
The first direct evidence of Policai Colla-do’s involvement in the conspiracy consists of two telephone calls between him and Frias that were recorded on September 21, 1988. The first direct evidence of Antonio Collado’s involvement in the conspiracy is found in a September 24, 1988 telephone call between Policai Collado and Frias in which Antonio Collado is mentioned.
The government points out that the two calls between Policai Collado and Frias were intercepted on the first day of the telephone surveillance. Moreover, it argues that a pre-existing relationship between the Collados and Frias can be inferred from these and other calls. Although the government does not press its point regarding the first three transactions,9 it contends that the inferences drawn from the calls intercepted on and after September 24, 1988 are sufficient to show that the Collados’ membership in the conspiracy reached back at least as far as September 13, 1988.
The district court made no finding on this issue. The presentence investigation report stated that the intercepted phone calls imply that the Collados were members of the conspiracy before the telephone surveillance began. The report also stated, however, that “it is unknown how long the defendants were] actually supplying the organization.”
We agree that the transcripts of the intercepted calls may be interpreted to mean that the Collados had transacted drug deals with Frias before September 21, 1988. We have no basis for deciding exactly when those prior transactions occurred, however, and thus, on this record, we cannot say whether Antonio Collado had joined the conspiracy by the time his brother made the disputed sale to Frias, or whether either brother had joined by the time Detective Chavez made the challenged purchases. Before the amounts involved in these transactions may be attributed to the Collados, the district court must find, by a preponderance of the evidence, United States v. Reyes, 930 F.2d 310, 314-15 (3d Cir.1991), that they were members of the conspiracy at the time the transactions occurred. Because the district court made no finding on this issue, we will vacate the judgment and remand for further fact finding.
The government’s brief, although not crystal clear on this point, could also be read as arguing that it would be appropriate to attribute to the Collados amounts distributed by their co-conspirators before they joined the conspiracy. Edwards stated that the defendant’s sentence may include amounts distributed before the defendant joined the conspiracy, provided that those transactions were reasonably foreseeable by the defendant and in furtherance of the activity he agreed to undertake. 945 F.2d at 1396-97. See also United States v. Matthews, 942 F.2d 779, 784 (10th Cir.1991); United States v. Miranda-Ortiz, 926 F.2d 172, 178 (2d Cir.), cert. denied, — U.S. -, 112 S.Ct. 347, 116 L.Ed.2d 287 (1991). In contrast, the Ninth Circuit recently held that accomplice attribution is appropriate only when the facts “show that the defendant aided and abetted [earlier] sales or was a member of the conspiracy” when those sales occurred. United States v. Chavez-Gutierrez, 961 F.2d 1476, 1479 (9th Cir.1992).
*997We note that the Edwards court’s position was influenced by that court’s interpretation of the law of conspiracy. As the Edwards court noted, 945 F.2d at 1395, under the law of that circuit, a defendant may be held liable for substantive offenses committed by a co-conspirator before the defendant joined the conspiracy. See, for example, United States v. Castillo, 814 F.2d 351, 355 (7th Cir.1987).
Other federal appellate courts disagree with the Seventh Circuit’s interpretation of conspiracy law. See United States v. Blackmon, 839 F.2d 900, 908-09 (2d Cir.1988); United States v. Carrascal-Olivera, 755 F.2d 1446, 1451 (11th Cir.1985). We need not take a position regarding the proper interpretation of conspiracy law, for in our view, the relevant conduct provision is not coextensive with conspiracy law.10 Wilkins and Steer, 41 S.C.L.Rev. at 510. The relevant conduct provision limits accomplice attribution to conduct committed in furtherance of the activity the defendant agreed to undertake. In the absence of unusual circumstances, not present here, conduct that occurred before the defendant entered into an agreement cannot be said to be in furtherance of or within the scope of that agreement.
B. Attribution to Each Brother of Amounts Supplied to the Conspiracy by the Other
In the letter memorandum he submitted after oral argument, see note 9, Antonio Collado concedes that he made a 125 gram sale of heroin to Frias sometime between September 29 and October 1, 1988 and a 187 gram sale sometime around October 10, 1988. Policai Collado argues that these amounts should not be attributed to him because “there is no evidence establishing that [he] was involved in, knew about or was otherwise responsible for these sales”. For the same reason, Antonio Collado argues that a 187 gram sale that Policai Collado admits making sometime around September 21, 1988 should not be attributed to him.11
The district court made no findings regarding the propriety of attributing to one brother sales made by the other, but after reviewing the transcripts of the telephone calls cited in the presentence investigation report, we are convinced that this instance of accomplice attribution was justified.The transcripts reveal that not only were the brothers aware of each other’s transactions with Frias, they also assisted each other, to some extent, in those transactions. For example, during a call recorded on October 20, 1988, Policai Collado told Frias that Antonio “ha[d] everything ready” for a scheduled drug deal and suggested that Frias call him back “so I can tell you at what time they left.” Moreover, Rosa German testified that she accompanied Frias to the apartment the Collados shared to obtain heroin and that on at least one of those visits, both brothers were present. We conclude that this evidence demonstrates that the transactions conducted by each brother were within the scope of the other’s agreement. See Garcia, 954 F.2d at 16-17 (evidence showed that co-conspirator’s drug sales were within scope of defendant’s agreement). Accordingly, we hold that it was appropriate to attribute to each brother amounts in*998volved in transactions conducted by the other.
C. Attribution to Both Brothers of Amounts Supplied to the Conspiracy by Third Parties
The Collados also contest attribution to them of the following three quantities: (1) the 125 grams that Detective Chavez purchased from Buena Ventura Marte; (2) the 16 grams seized from Ramon Rivera upon his arrest; and (3) the 125 grams discussed in an intercepted phone call between Frias and Alberto Herrera, a member of the Philadelphia organization. The Collados contend that because each of these three amounts was supplied to the conspiracy by other persons, it was error to attribute the amounts to them.12
The factual findings of the district court are not sufficient for us to resolve this issue. In the absence of factual findings on the scope of the Collados’ agreement with their co-conspirators, the reasonable foreseeability to them of the transactions, and the degree of their involvement in the conspiracy, we cannot determine whether accomplice attribution of these amounts is justified. Accordingly, we will remand to the district court for further fact finding.
V. THE SUFFICIENCY CHALLENGE
The Collados also challenge the government’s estimate of the quantities involved in two specific transactions.13 We recognize that in calculating the amounts involved in drug transactions, some degree of estimation must be permitted, for the government usually cannot seize and measure all the drugs that flow through a large drug distribution conspiracy. This is not to say that calculations of drug amounts may be based on mere speculation, however. Given the dramatic effect such estimates have on the defendant’s sentence, the sentencing court must carefully review the government’s submissions to ensure that its estimates are proven by a preponderance of the evidence. As noted at page 990, we review the district court’s calculations only for clear error.
The government estimated the amount of heroin involved in the first transaction at 62.5 grams. The evidence presented regarding this transaction consisted of transcripts of two telephone calls taped on October 20, 1988. The first call was placed by Frias to Policai Collado; the second was placed by Angito Sanchez, a member of the Philadelphia organization, to Antonio Collado. The government’s expert witness testified that both calls referred to a transaction in which the Collados agreed to supply heroin to Frias.
Neither call contains any reference to a specific amount of heroin. The presen-tence investigation report observes that “based on other telephone calls related to this transaction, the government stated that this conversation is referring to 62.5 grams of heroin.” As the defendants point out, the report does not explain what those other calls were, nor did any trial testimony establish that these calls referred to 62.5 grams of heroin. The government’s only argument is that the calls must have referred to a sale of at least 62.5 grams because the Collados always dealt in quantities of at least that amount. In our view, the government’s argument crosses the line between estimation and speculation. Because there is no evidentiary basis for concluding that this transaction involved 62.5 grams of heroin, we conclude that it *999was clear error to include this amount in the calculation of the defendants’ base offense levels.
The government estimated the amount involved in the second transaction as 125 grams.14 To support this estimate, the presentence investigation report relies upon a conversation that was recorded on November 1, 1988, but was not introduced at trial. The speakers, identified as Pedro Frias and Alberto Herrera, state in relevant part:
PEF: To check and see which is the better one.
AH: Umhum.
PEF: The one that arrived yesterday is the one that he worked the day before. AH: Umhum.
PEF: Umhum, alright, then, I’ll tell the people.
AH: Alright, I’ll tell him to bring of [sic] the one from yesterday, yes.
PEF: The one from yesterday, to work that one, because that’s the one that he worked last week.
AH: Umhum.
PEF: And he worked 2 quickly, he did. AH: Yes.
PEF: Yes, so, to work that one, the, if he sees that it’s giving results, to mix it already, the little bit that was left from yesterday.
AH: No, because one from hasn’t been started. There are 2, there are 2 that haven’t been started.
PEF: Start yesterday’s, that one is good, that’s the one that was sold the day before.
Although defendants contend that “one” should be read in the generic sense, the government presented ample evidence indicating that the conspirators used the term to refer to specific quantities of drugs. The government’s expert witness, DEA Agent Hershowitz, testified at several points that the conspirators used the term “one” to refer to one-eighth of a kilo of heroin, or 125 grams. Moreover, German also testified that members of the conspiracy used the term “one” as a code word for specific quantities of drugs. In light of this evidence, we conclude that the district court’s acceptance of the government’s estimate of the quantity involved in this transaction was not clearly erroneous.
VI. CONCLUSION
For the foregoing reasons, we will vacate the judgment and remand for further sentencing proceedings consistent with this opinion.
. We refer to Frias as an "alleged” manager because he was a fugitive at the time of the Collados’ trial and, as far as we know, remains a fugitive today.
. As far as we know, Rivera also remains a fugitive.
.The Guidelines provide that the base offense level corresponding to a quantity of heroin between 700 and 1000 grams is 30. See Drug Quantity Table, U.S.S.G. § 2Dl.l(c).
. One offense — simple possession of a controlled substance other than cocaine base weighing more than 5 grams — is treated differently under the Guidelines. Under U.S.S.G. § 2D2.1, the base offense level for that crime depends upon the type of drug involved. Even in such cases, however, the sentencing court may depart upward based on drug quantity. See, for example, United States v. Ryan, 866 F.2d 604 (3d Cir.1989).
. The relevant conduct provision also describes a second method, not at issue here, of including amounts in addition to those amounts the defendant was convicted of distributing: it authorizes the inclusion of amounts distributed by the defendant himself or herself, but for which he or she was not convicted or charged.
. Although this court has reviewed the application of the relevant conduct provision, we have not been required previously to explain in detail the legal standard for accomplice attribution. In United States v. Salmon, 944 F.2d 1106 (3d Cir.1991), cert. denied as Washington v. United States, — U.S.-, 112 S.Ct. 1213, 117 L.Ed.2d 451 (1992), for example, we relied upon the relevant conduct provision to attribute to defendant amounts involved in transactions at which defendant was present. In United States v. Headley, 923 F.2d 1079 (3d Cir.1991), the defendant had been sentenced based on the entire amount of drugs distributed by the conspiracy, even though her involvement in the conspiracy may have been limited to acting as a courier. She raised no objection to the propriety of accomplice attribution at the trial or appellate level, however. Rather, on appeal she argued that her trial counsel’s failure to .request that the sentencing court adjust her base offense level to reflect a minor or minimal role in the offense constituted ineffective assistance of counsel. We agreed, and remanded for resentencing.
United States v. Gonzalez, 918 F.2d 1129 (3d Cir.1990), cert. denied, — U.S.-, 111 S.Ct. 1637, 113 L.Ed.2d 733 (1991), and United States v. Williams, 917 F.2d 112 (3d Cir.1990), cert. denied, — U.S. -, 111 S.Ct. 1001, 112 L.Ed.2d 1084 (1991), did not involve amounts of drugs distributed by accomplices; instead, both cases involved amounts distributed by the defendants, but of which they were not convicted of distributing. Moreover, in Williams, the defendant stipulated to the appropriate base offense level.
. The court explained its reference to these two factors by stating that it was looking to conspiracy law for guidance in considering which conduct is relevant in sentencing. 945 F.2d at 1392. The leading case of Pinkerton v. United States, 328 U.S. 640, 66 S.Ct. 1180, 90 L.Ed. 1489 (1946), established that a defendant is responsible for a co-conspirator’s acts when those acts are (1) committed in furtherance of the conspiracy and (2) reasonably foreseeable by the defendant as a necessary or natural consequence of the unlawful agreement. 328 U.S. at 646-47, 66 S.Ct. at 1183-84.
. Our interpretation of the relevant conduct provision differs from that in Edwards in one significant respect. Edwards states that accomplice conduct committed before the defendant joined the conspiracy may be attributed to the defendant. As we explain at pages 997-998, in our view, such attribution is justified only in unusual circumstances.
. After oral argument, pursuant to the court’s request, the government submitted a letter memorandum (to which the defendants responded) detailing the amounts it contended should be attributed to the Collados. The government’s memorandum did not include the three earlier transactions, although the district court attributed the amounts involved in those transactions to the Collados.
. For the same reason, we need not decide whether the standard for accomplice attribution described in the relevant conduct provision is the same as or narrower than the Pinkerton standard. Compare United States v. Perrone, 936 F.2d 1403, 1416 (2d Cir.1991) (“scope of conduct for which a defendant can be held accountable under the sentencing guidelines is significantly narrower than the conduct embraced by the law of conspiracy”), clarified on other grounds, 949 F.2d 36 (2d Cir.1991), and Wilkins & Steer, 41 S.C.L.Rev. at 510 (relevant conduct provision narrower than Pinkerton), with Edwards, 945 F.2d at 1392 (standards the same); United States v. Joyner, 924 F.2d 454, 458 (2d Cir.1991) (same); and United States v. La-Fraugh, 893 F.2d 314, 317 (11th Cir.) (same), cert. denied, 496 U.S. 911, 110 S.Ct. 2601, 110 L.Ed.2d 281 (1990).
. Antonio Collado also contends that the government failed to prove that he was a member of the conspiracy at the time this sale was made, an argument we addressed at pages 996-998. In this section of our analysis, we assume without deciding that Antonio Collado was a member of the conspiracy when Policai Collado made this sale.
. We cannot determine the source of the second two quantities on the basis of this record. We assume, for the purposes of our analysis here, that these amounts were supplied by persons other than the Collados. We do not intend our assumption to foreclose the district court from reaching a different conclusion upon remand.
. In addition, the defendants challenge the probation officer's overall calculation of the amount of drugs involved in the conspiracy, contending that it is inconsistent with Rosa German's plea agreement, which stated that 400 to 700 grams were involved. We find this challenge meritless. The district court was not bound by German’s plea agreement in sentencing the Collados. Moreover, the government has explained that, by the time the Collados were sentenced, it had more information about the conspiracy than it did when German entered her plea.
. In addition to contesting the amount involved in this transaction, the Collados also contend that it was error to attribute the transaction to them because there was no evidence to show that they supplied the amount involved. We discuss that aspect of their argument at pages 996-97.