Naturalness and effectiveness of gesture-based communication strongly depend on the success of gesture recognition. However, confusion in classification increases when considering gestures with similar evolutions. Given that neither typical motion-based features, nor hidden Markov models are capable to distinguish accurately among them, it is common to consider only gestures that require different forms of execution. In this paper, we present empirical evidence showing that, in addition to motion, posture information significantly increases classification rates, even with similar gestures. Moreover, for recognition, we propose dynamic naive Bayesian classifiers. In comparison to hidden Markov models, these models require less iterations of the EM algorithm for training, while keeping competitive classification rates. The proposed system was evaluated considering 9 classes of similar gestures, showing a significant increase in performance by integrating motion and posture attributes.