Four hundred and sixty-eight class-II amalgam restorations were placed in 210 patients. The restorations had been inserted by seven Scandinavian dentists in their clinics, among their regularly attending patients. Impressions of the teeth with cavity preparations had been made, and epoxy casts fabricated. The designs and qualities of the cavities were assessed in accordance with an evaluation system for class-II cavities designed for use on models. The restorations were observed yearly and scored in accordance with the USPHS criteria. In case of replacement the reason was recorded, and the service time calculated. The restorations were observed throughout a period varying from 8 to 10 years. At the end of the observation period, 212 restorations had been lost owing to dropout patients, 68 restorations had been replaced, and 188 restorations remained functional. The commonest criteria for replacement were secondary caries (n = 30) and restoration bulk fractures (n = 24). Using univariate statistics and multivariate discriminant analyses, the time of service and the reasons for replacement were correlated with different clinical variables, including different indices for the dimensions and qualities of the cavity preparation. Several features of the cavity design could be associated with the service time of the restoration or with the reason for replacement, or both. Secondary caries was primarily associated with cavity design features gingivally on the proximal surface (p less than 0.001). At the patient level the rate of secondary caries correlated with the total number of restorations placed during the observation period, irrespective of the quality of the cavosurface margins or the size of the cavity. Restoration bulk fractures could be related to cavities with narrow and deep occlusal parts, or deep proximal parts (p less than 0.001). The discriminant functions predicted correctly between 70% and 93% of the actually failed restorations in the failure groups. The good prediction performance indicates that a linear discriminant analysis that includes aspects of the prepared cavity may be applied to predict the reasons for failure of restorations.