Objective: To examine the occurrence of critical incidents (CIs) in order to improve quality of care.¶Design: Prospective survey.¶Setting: Multidisciplinary, neonatal-pediatric intensive care unit (ICU) of a non-university, teaching children's hospital.¶Patients: Four hundred and sixty-seven admissions over a 1-year period.¶Methods: A CI is any event which could have reduced, or did reduce, the safety margin for the patient. Comprehensive, anonymous, non-punitive CI monitoring was undertaken. CI severity with respect to actual patient harm was graded: major (score 3), moderate (2) or minor (1). The system approach incorporates the philosophy that errors are evidence of deficiencies in systems, not in people. We undertook 2-monthly analyses of CIs.¶Results: There were 211 CI reports: 30 % major, 25 % moderate, 45 % minor. The CI categories were management/environment 29 %, drugs 29 %, procedures 18 %, respiration 14 %, equipment dysfunction 7 %, nosocomial infections 3 %. The respiratory CIs were the most severe, the drug-related CIs the least severe (score mean, SD: 2.9, 0.26 vs 1.4, 0.76; p < 0.001). However, 20 out of 62 drug-related CIs were potentially life-threatening. Thirteen percent of drug CIs were decimal point errors. Eleven of the 29 respiratory CIs were accidental extubations (2.6/100 ventilator days). CIs were most often precipitated by consultants (32 %), followed by residents (23 %, over-represented in drug CIs, 22/62) and specialized nurses (21 %). Doctors had a greater proportion of major CIs than nurses (p < 0.01). Fifty percent of the CIs were detected by routine checks. The most important method of detection was patient inspection (44 %), alarms accounted for only 10 %. Contributing factors were human errors (63 %), communication failure (14 %), organizational problems (10 %), equipment dysfunction (7 %) and milieu (3 %).¶Conclusion: CIs are very common in pediatric intensive care. Knowledge of them is a precious source for quality improvement through changes in the system.