This study was based on data collected from 87 men during and immediat
ely following participation in a counseling program aimed at assisting
them to reduce sexual behaviors associated with high HN risk. All the
melt reported having engaged in same-gender anal or oral sex without
condoms during the year prior to enrollment. Three-month retrospective
reports of sexual behavior, solicited just after participation, were
compared with reports collected weekly during the same three-month per
iod. Accuracy of recall over an ensemble of 10 behavior items was quit
e low with amount of error positively associated with behavior frequen
cy. In general, exaggerated reports were associated with low-frequency
behavior and underreports with high-frequency behavior. Because of ob
served differences in the average frequency of occurrence of the vario
us specific types of sexual behavior, adjustment for event frequency w
as required to provide adequate analysis of between-behavior differenc
es in memory error. Estimated functional relationships between behavio
r frequency and average memory error illustrated that for sexual activ
ity occurring between 1 and 20 times in the three-month period, there
was a significant association, between the HIV-risk level of a behavio
r and the average extent to which the behavior was forgotten (i.e., un
derreported). After event-frequency adjustment, anal activity without
condoms-the behavior believed to entail the highest level of risk for
HN transmission, and the behavior about which the counseling program p
rovided the least ambiguous cues regarding risk-was associated with th
e greatest underreporting error, and the ensemble of unprotected behav
iors (i.e., those unaccompanied by condom usage) involved significantl
y greater underreporting than. did protected behaviors. Error rates ov
er the ensemble of behaviors were strikingly Low for zero-frequency ev
ents, suggesting the utility of dichotomous ever/never measures in res
earch evaluating activity levels and behavior change. Other analysis t
echniques for reducing potential recall bias, allowing finer evaluatio
ns of behavior frequency and change, were also discussed.