This paper reports experiments with the generalized binary probabilistic independence model in which more complete statistical information is used. Two basic sets of experiments, referred to as nonpredictive and predictive, have been performed on two standard test collections. In the nonpredictive experiments, the complete relevance information was used to determine the optimal performance of the model. On the other hand, in the predictive experiments only partial relevance information was used to demonstrate the predictive power of the model. Although a simple method for estimating the parameters was used in the designed tests, significant improvements were obtained for both test collections. These preliminary results suggest that further work on the generalized model is worthwhile.