OBJECTIVE: To assess the ability of computed tomography in predicting whether suspected ovarian cancer could be fully excised at primary laparotomy. DESIGN: Retrospective analysis of patient notes and pre-operative computed tomography scans. Setting A UK NHS cancer centre. POPULATION: Seventy-seven women who underwent laparotomy for an ovarian tumour and who had had a pre-operative computed tomography scan. METHODS: Women who had a computed tomography scan before laparotomy for an ovarian tumour were identified. Analysis was undertaken to determine the accuracy of computed tomography in predicting malignancy, stage and residual disease. The computed tomography parameters significantly associated with residual disease were determined by a chi2 analysis. These parameters, in addition to age and CA125, were used to generate a predictive model. This model was further refined by stepwise logistic regression and a clinical scoring index was generated. MAIN OUTCOME MEASURES: To identify those computed tomography parameters significantly associated with residual disease and to use these with CA125 and age to generate a useful clinical scoring index to predict residual disease in suspected ovarian cancer. RESULTS: Seventy-seven women underwent a laparotomy for an ovarian tumour and had a pre-operative computed tomography scan. Fifty-one of these women had malignant disease and twenty-five of these women had residual disease remaining. The sensitivity of computed tomography in predicting malignancy was 90% with a specificity of 85% and the overall accuracy of computed tomography for predicting stage of disease was 73% (37/51). The overall sensitivity of computed tomography in predicting residual disease was 88%, the specificity was 92% and the positive predictive value was 85%. The parameters on computed tomography that were significantly (P <0.05) associated with residual disease were ascites, omental cake, mesenteric disease, paracolic gutter deposits, diaphragmatic deposits and pleural effusion. The predictive model generated was more accurate than computed tomography alone (sensitivity 88%, specificity 98%, positive predictive value 95%). Using stepwise logistic regression enabled the predictive model to be simplified to include mesenteric disease, omental cake, age and CA125 without any change in sensitivity or specificity and this model was used to generate a scoring index. CONCLUSION: This study shows that prediction of resectability by computed tomography is excellent and is further improved by the generation of a predictive model, which can be used to generate a simple scoring index. This scoring system now needs to be tested prospectively to ensure that its performance remains as good in an independent sample population.