This talk describes and empirically evaluates a Quasi-Incremental Bayesian Classifier (QBC) designed to be used when a classification task must be performed in dynamic systems such as sensor networks, which are continuously receiving new piece of information to be stored in huge databases. Therefore, the knowledge that needs to be extracted from these databases is continuously evolving and the learning process may need to go on almost indefinitely. The induction proposed by QBC is performed in two steps; in the first one a traditional Bayesian Network (BN) induction algorithm is performed using an initial amount of data. As far as new data is available, only the numerical parameters of the classifier are updated. The conducted experiments showed that QBC tends to maintain the average correct classification rates obtained with non-incremental classifiers while decreasing the time needed to induce the classifier.