Research on adverse outcomes following common surgical procedures has suggested the importance of hospital and surgeon variables. Policy directions depend on which factors are important in influencing patient outcomes and what sorts of policies are feasible. Focusing on where a given procedure is performed highlights a concern for centralization; emphasizing who should perform a particular operation implies physician certification. Finally, monitoring involves identifying particular hospitals that appear to have relatively poor (or relatively good) results. This paper analyzes patient, surgeon, and hospital characteristics associated with serious postdischarge complications of hysterectomy, cholecystectomy, and prostatectomy in patients age 25 and over in Manitoba, Canada, following surgery during 1974 through 1976. The three procedures differ markedly in the ease of prediction of the probability of complications and in the predictive importance of patient, hospital, and physician variables. The predictors worked fairly well for cholecystectomy, somewhat less well for hysterectomy, and not well at all for prostatectomy. Hospital variables were not generally important in the multiple logistic regressions. After controlling for case mix and type of surgery, physician surgical experience was found to account for relatively large differences (almost two to one) in the probability of patient complications following cholecystectomy. Cholecystectomy might be a candidate for certification because of the epidemiology of the operation. As of the mid-1970s, a substantial proportion of the cholecystectomies were being performed by physicians with comparatively little ongoing experience with this type of procedure. Moreover, a monitoring perspective identified one hospital with a significantly higher postcholecystectomy complication rate, even after physician experience was taken into account. Both identifying which procedures should be attended to and focusing on problems following surgery are important beyond Manitoba and highly relevant to such American requirements as Peer Review Organizations. Methods of increasing the efficiency of using claims data for quality assurance studies are outlined.