OBJECTIVES: Previous studies have found social cognitive theory (SCT)-framed interventions are successful for improving condom use and reducing sexually transmitted infections (STIs). We conducted a secondary analysis of behavioural data from the Safe in the City intervention trial (2003-2005) to investigate the influence of SCT constructs on study participants' self-reported use of condoms at last intercourse. METHODS: The main trial was conducted from 2003 to 2005 at three public US STI clinics. Patients (n=38,635) were either shown a 'safer sex' video in the waiting room, or received the standard waiting room experience, based on their visit date. A nested behavioural assessment was administered to a subsample of study participants following their index clinic visit and again at 3 months follow-up. We used multivariable modified Poisson regression models to examine the relationships among SCT constructs (sexual self-efficacy, self-control self-efficacy, self-efficacy with most recent partner, hedonistic outcome expectancies and partner expected outcomes) and self-reported condom use at last sex act at the 3-month follow-up study visit. RESULTS: Of 1252 participants included in analysis, 39% reported using a condom at last sex act. Male gender, homosexual orientation and single status were significant correlates of condom use. Both unadjusted and adjusted models indicate that sexual self-efficacy (adjusted relative risk (RRa)=1.50, 95% CI 1.23 to 1.84), self-control self-efficacy (RRa=1.67, 95% CI 1.37 to 2.04), self-efficacy with most recent partner (RRa=2.56, 95% CI 2.01 to 3.27), more favourable hedonistic outcome expectancies (RRa=1.83, 95% CI 1.54 to 2.17) and more favourable partner expected outcomes (RRa=9.74, 95% CI 3.21 to 29.57) were significantly associated with condom use at last sex act. CONCLUSIONS: Social cognitive skills, such as self-efficacy and partner expected outcomes, are an important aspect of condom use behaviour. TRIAL REGISTRATION NUMBER: clinicaltrials.gov (NCT00137370).