A lot of researches have been done, so far. Among them, it can be mentioned to using finite element method, correction of Mixture rule or micromechanical theories for unidirectional short fiber composites. Bear in mind that some of these theories have limiting preconditions and in some others the effects of fiber length and fiber orientation distribution are not considered, in order to develop an efficient model which is able to predict stiffness of short fiber composites, ANN approach is applied in current study. The main advantage of this method over previous methods is that there aren’t any limiting preconditions in it. In order to train ANN, these steps have been followed: production of polypropylene reinforced with short glass fibers, measurement of fiber length and fiber orientation distribution through image analysis, measurement of longitudinal elastic modulus of the produced composites, training a variety of ANN with different structures to find the best network which can do the prediction most efficiently, applying four different models into obtained data and evaluation the accuracy of these models and the model created using ANN. Final step has been done through comparison the results of calculated modulus by the models with experimental results to examine the agreement between them. Based on the results obtained in this study, paper physics approach (PPA) can predict longitudinal elastic modulus more efficiently than others. Laminate analogy approach (LAA) and ANN get the second rank, rule of thumb and Cox-Krenchel methods get the third and fourth rank, respectively. These facts are revealed through root mean square error (RMSE) which is calculated for each model. The values of RMSE are 0.7206, 1.0673, 1.1331, 1.2643 and 1.4230 for PPA, LAA, ANN, rule of thumb and Cox-Krenchel, respectively. So the best model for prediction of longitudinal modulus of short fiber composites is PPA. It is also found that rule of thumb is a suitable and simple method for estimation of longitudinal modulus of short fiber composites with low amount of fiber, for example 10 or 20 percent of weight fraction. The main result of this research is that ANN is an appropriate approach to predict stiffness of short fiber composites. This is proved by the model created using ANN.