Using a comprehensive anesthesia simulation environment (CASE 1.2) we studied the response of anesthesia trainees (10 first-year residents and 9 second-year residents) to five simulated critical incidents: 1) endobronchial intubation; 2) kinked IV; 3) atrial fibrillation with hypotension; 4) breathing circuit disconnection; 5) cardiac arrest. Simulations were videotaped, and the response times for detecting and initiating correction of the problems were measured. Different problems had significantly different response characteristics. Breathing circuit disconnection and cardiac arrest were quickly detected (21 +/- 17 seconds; 7 +/- 5 seconds), and correction was begun quickly, although major errors in managing the cardiac arrest occurred in 58% of cases. Endobronchial intubation and atrial fibrillation took longer than the other problems to detect (105 +/- 142 seconds; 111 +/- 158 seconds) and to correct (555 +/- 358 seconds; 365 +/- 121 seconds). Intravenous kink was detected more slowly (238 +/- 269 seconds) but once discovered was quickly corrected. The response of different individuals was highly variable; experience level was a significant factor for correction (P = 0.03) but not for detection of problems overall. Because of high individual variation, experience was not a significant factor in correcting any signal problem. The data suggest that some types of problems are harder to handle than others and that individuals vary widely in their problem-handling abilities. Experience is a beneficial factor in anesthesia problem solving but not in a simple fashion. Vigilant detection of problems is only the first step in a complex response pathway that might be strengthened by improved protocols and repeated practice.