Miniature disk bending test was developed to evaluate the mechanical behavior of irradiated materials - mainly to identify the ductility loss in steels. The analytical solution for large amplitude, elasto-plastic deformation with contact and friction becomes rather unwieldy. It is difficult to distinguish between the regimes of elastic and plastic deformation since local plastic deformation occurs for very small values of load when the magnitude of spatial average stress is well below the yield stress. Thus the inverse problem of evaluating properties from the experimentally observed values of the deflection of the specimen is rather difficult to solve analytically. The paper discusses some of the published work in this area and the difficulties associated with them. The approach in this work is to first generate a large database - by a finite element (FE) solution - of load-displacement (P-w) records for varying material properties viz. elastic modulus (E), yield stress (σy) and the constants (A0, m) appearing in the plastic stress-strain relation. An artificial neural network (ANN) is trained with the specified material properties and the corresponding load-displacement records. In the second phase, this network is tested with known data and then used with the experimentally observed (P-w) records to predict the abovementioned material properties. The paper presents the details of modeling (FE and ANN), a summary of the results obtained by the FE analysis (database) and the results obtained by the ANNs in the training and the testing phases. The errors in the various values of the parameters during testing were found to be within 5%.