Qualitative insights about students' personal experience of inconsistencies in implementation of problem-based learning (PBL) might help refocus expert discourse about good practice. Aim: This study explored how junior medical students conceptualize: PBL; good tutoring; and less effective sessions. Methods: Participants comprised junior medical students in Liverpool 5-year problem-based, community-orientated curriculum. Data collection and analysis were mostly cross-sectional, using inductive analysis of qualitative data from four brief questionnaires and a "mixed" qualitative/quantitative approach to data handling. The 1999 cohort (end-Year 1) explored PBL, generated "good tutor" themes, and identified PBL (dis)advantages (end-Year 1 then mid-Year 3). The 2001 cohort (start-Year 1) described critical incidents, and subsequently (end-Year 1) factors in less effective sessions. These factors were coded using coding-frames generated from the answers about critical incidents and "good tutoring". Results: Overall, 61.2% (137), 77.9% (159), 71.0% (201), and 71.0% (198) responded to the four surveys, respectively. Responders perceived PBL as essentially process-orientated, focused on small-groupwork/dynamics and testing understanding through discussion. They described "good tutors" as knowing when and how to intervene without dominating (51.1%). In longitudinal data (end-Year 1 to mid-Year 3), the main perceived disadvantage remained lack of "syllabus" (and related uncertainty). For less effective sessions (end-Year 1), tutor transgressions reflected unfulfilled expectations of good tutors, mostly intervening poorly (42.6% of responders). Student transgressions reflected the critical incident themes, mostly students' own lack of work/preparation (54.8%) and other students participating poorly (33.7%) or dominating/being self-centred (31.6%). Conclusion: Compelling individual accounts of uncomfortable PBL experiences should inform improvements in implementation.