In this paper a novel unconstrained numeral recognition system using MFNN based multiexpert combination strategy is proposed. Compared with many traditional methods, the proposed approach takes advantage of every method*s confidence value as well as its output label. So it can improve the system*s performance significantly with as few as two experts. Another novel feature of the system is that the multiexpert combination problem is converted to a new classification problem and then MFNNs are used to handle the combination adaptively. The proposed system have achieved promising result on the NIST database. When rejection rate is zero, its recognition rate is 98.24% and the error rate can be reduced to 0.24% with a rejection rate of 5.72%.