In this work a method for splitting continuous mixture density hidden Markov models (HMM) is presented. The approach com-bines a model evaluation measure based on the Maximum Mutual Information (MMI) criterion with subsequent standard Max-imum Likelihood (ML) training of the HMMparameters. Experiments were performed on the SieTill corpus for telephone line recorded German continuous digit strings. The proposed split-ting approach performed better than discriminative training with conventional splitting and as good as discriminative training after the new splitting approach.