The traditional method of design for injection molded parts does not account for the variation of mechanical properties within a part. Knowledge of part properties before actual molding and testing will be of immense help to part and mold designers in modification of design. Thus many methodologies to predict mechanical properties in injection molded parts have been developed. In this paper, the methodology involved the application of connection learning system, injection molding computer simulation, and experimental evaluation of mechanical properties, to relate the thermomechanical history to the resulting part properties so that they can be predicted before parts are molded. The mechanical properties of injection molded parts depend upon their thermomechanical history which in turn is greatly influenced by the processing condition and part geometry As the relationship between mechanical properties and thermomechanical history is complex and highly non-ljnear, the developed methodology was based on a backpropagation neural network that provided the means for a non-parametric mapping between them. When thermomechanical history can be obtained from injection molding simulation, this methodology Provides designers with the ability to Predict mechanical properties in injection molded parts of another processing condition. The main weakness of backpropagation is that it cannot escape the local minimum. This paper applied the momentum parameter and noise factor, and thus although the some initial value was given in backpropagation, the predicted result was in good agreement with the mechanical properties.