In this paper we present an adaptive classification method that features a robust, efficient and simple to use incremental clustering algorithm. A new assignment strategy for incorporating new data patterns allows clusters to align more exhaustively with the data structure. This almost eliminates the sensitivity to the order of input data, many incremental clustering algorithms suffer from, reduces the number of clusters needed and thus improves also time efficiency. For updating the clusters? representations we utilize an incremental version of PCA which generates its learning rate automatically from the number of patterns. Furthermore, the size and number of clusters is controlled by the classification error. So we get a classification method where nothing but the target error needs to be pre-specified. We conducted experiments on artificial and real data to demonstrate the capabilities of the proposed algorithm.