The paper contains a report of a test of Cooper's NewProd model for predicting success and failure of product development projects. Based on Canadian data, the model has been shown to make predictions which are 84% correct. Having reservations on the reliability and validity of the model on theoretical grounds, we set out to test it using data collected in the Netherlands. Following Cooper's methodology, we selected 19 projects, which had already been marketed and of which 9 were clearcut successes and 10 clearcut failures. We also studied 9 additional projects, which had not yet been commercialised, as the basis for a future a priori test. The projects were given a product score according to Cooper's criteria, and predictions compared with the actual experience. Eightyfour percent were correctly classified, as in Cooper's work. From a practical point of view however the variance was too large to allow the predictions from NewProd to be used with confidence to predict outcomes unless they are clearcut. We believe that the drawback of the methodology is that the product score is a simple combination of the various factors involved. No allowance is made for the possibility that one factor alone might be responsible for failure. We have therefore introduced the concept of a threshold value for each factor. A project for which any factor falls below this value will be deemed a failure. Based on a very limited sample of 19 cases an improvement of reliability was made to 95% of 18 out of 19 products correctly classified. The NewProd concept so modified is being applied to the sample of 9 as yet uncompleted projects, and its predictions will eventually be compared with actual outcomes.