Multivariate models have great potential value in enhancing the understanding of why some pregnancies have poor outcomes. Recently, such models have been advocated as a basis for predictive scoring systems that attempt to classify patients into high-risk and low-risk groups. In this report the usefulness of such an approach was assessed by studying the predictability of preterm delivery at The Johns Hopkins Hospital during 1980, using a multiple logistic model. Choosing a cutoff point (or probability of preterm delivery) of 10%, 697 of 2865 patients were placed in the high-risk group. The sensitivity, specificity, and positive predictive value of the model, as applied to this select population, were 62.2, 79.4, and 22.7%, respectively. Thus, only 23% of patients predicted to have preterm deliveries in fact delivered preterm. The predictive value could have been improved by increasing the cutoff point, but only at the expense of markedly reducing the sensitivity of the model. It was concluded that the potential value of multivariate analyses of pregnancy outcome as a predictive, risk-classification technique is limited. Nevertheless, such studies may aid the clinical evaluation of each individual patient by providing a better understanding of the etiologies of poor outcome.