Abstract: This paper deals with function approximation capabilities of monolithic fuzzy neural networks （MFNNs）. In MFNNs the basic operators are Min and Max which replace the multiply and sum operators in traditional neural networks. This makes various differences in properties of approximation to function. Proposed in the paper are the ordered monotony property of MFNNs when the transfer function is monotone, the continuous mapping theorem, and non approximation theorem to function. It is shown that although MFNNs can keep continuous mapping property, their capabilities to approximate function is worse comparing with traditional neural networks.