To explore the relationship between self-directed learning (SDL) needs, patterns, barriers, and outcomes among nonacademic general surgeons. Participants dictated details of SDL episodes associated with cancer patient management from October 2007 to March 2008. Transcripts were coded thematically. Frequencies were calculated for elements of each SDL stage. Statistical significance among subgroups was established with the use of the Pearson chi-square test, adjusted for clustering by surgeon. Participants were interviewed by telephone, and transcripts were analyzed by qualitative methods. Of 21 consenting surgeons, 15 submitted 115 cases, and 108 were analyzed. Most involved breast (40.7%), colon (18.5%), or rectal cancer (13.0%); 2 or more clinical tasks (41.7%); and 2 or more questions (89.8%). Information was sought from the Internet (48.1%), colleagues (24.2%), or both (6.8%). Information was partially, or not relevant for 21.3% of cases. Evidence was new for 66.7%, and confirmed knowledge for 10.7% of cases. Learning helped surgeons formulate new (34.2%), or confirm original (16.5%) management plans, or determine that referral was appropriate (39.2%). Use of codified sources was associated with information retrieval (P < .05), and identifying new evidence leading to a change in management from that initially proposed (P < or = .001). Numerous individual and systemic barriers may prevent practicing physicians from undertaking SDL, but provision of structured guidance prompted SDL and resulted in several beneficial outcomes. Further research is needed to validate these findings, and investigate who should support SDL, and how.