Underperforming trainees requiring remediation may threaten patient safety and are challenging for vocational training programs. Decisions to institute remediation are high-stakes--remediation being resource-intensive and emotionally demanding on trainees. Detection of underperformance requiring remediation is particularly problematic in general (family) practice. We sought to establish early-training assessment instruments predictive of general practice (GP) trainees' subsequently requiring formal remediation. We conducted a retrospective cohort study of trainees from a large Australian regionally-based GP training organization. The outcome factor was requirement for formal remediation. Independent variables were demographic factors and a range of formative assessments conducted immediately prior to or during early-stage training. Analyses employed univariate and multivariate logistic regression of each predictor assessment modality with the outcome, adjusting for potential confounders. Of 248 trainees, 26 (10.5%) required formal remediation. Performance on the Colleague Feedback Evaluation Tool (entailing feedback from a trainee's clinical colleagues on clinical performance, communication and probity) and External Clinical Teaching Visits (half-day sessions of the trainee's clinical consultations observed directly by an experienced GP), along with non-Australian primary medical qualification, were significantly associated with requiring remediation. There was a non-significant trend for association with performance on the Doctors Interpersonal Skills Questionnaire (patient feedback on interpersonal elements of the consultation). There were no significant associations with entry-selection scores or formative exam or assessment scores. Our finding that "in vivo" assessments of complex behaviour, but not "in vitro" knowledge-based assessments, predict need for remediation is consistent with theoretical understanding of the nature of remediation decision-making and should inform remediation practice in GP vocational training.