This paper studies and compares the effectiveness of four different features and their combinations on the recognition accuracy of handwritten Bangla characters. The longest run, chain code histogram, shadow, and Gabor filter based features and their eleven (11) combinations were tested on a standard Bangla database of 15, 000 basic handwritten characters to compare their recognition performances. From the experiments performed, it was observed that the combination of the longest run, chain code histogram, and the shadow features (having feature vector sizes of 20, 20, and 16 respectively) produce the highest recognition accuracy of 84.01%. Furthermore, inclusion of a feature with a large vector size compared to the other features in the combination generally dominates the recognition accuracy. In our case, inclusion of the Gabor filter-based features with a vector size of 1024 in the combination produced a recognition accuracy of 69.71%, which is worse than the accuracy obtained using the other three features. The analysis of the results indicates that the combinations of different feature vectors produce better accuracy as long as the sizes of each individual feature vector is comparable with each other in the combination.