Character recognition is an important task in biometrics. This paper uses neural network for real time handwritten Marathi numerals recognition. We have taken 150 online Marathi numerals written in different styles by 10 different persons. Out of these, 50 numerals were used for training purpose and another 100 numerals were used for recognition purpose. The numerals undergo the preprocessing steps using image processing techniques and after character digitization it is further subjected to the multilayer backward propagation neural network for recognition purpose. The proposed research work gives recognition accuracy from 97% and to 100% for the different resolution of input vector.