Nowadays a considerable quantity of metallic parts has been manufactured via metal forming processes, such as rolling, stamping, drawing, etc. at the ambient temperature. However, limited workability of some alloys may lead to shaping them at higher temperatures, where the material can flow more easily and the part would contain less defects. Since the stress at low temperature is a function of strain (the relative amount of deformation), the flow behavior of material at these temperature is rather straightforward. At high temperature other parameters, namely temperature and strain rates, should be also taken into account in predicting the flow stress. Our goal in this research is to predict the high temperature behavior by introducing different constitutive and numerical approaches. The Ti-6Al-4V alloy’s promising mechanical properties such as excellent strength-to-weight ratio and great chemical and thermal resistance make it a great choice to be applied in different industries such as medical, marine, aerospace. This Thesis focuses on predicting the behavior of this alloy under tension at high temperatures and aims to find a model which can describe the stress-strain relevance accurately under desired external circumstances required for any deformation process. There are various models trying to exploit the non-linear stress dependencies on temperature, strain, and strain rate. Herein, the constitutive and the Artificial Neural Network (ANN) models will be explained with all their benefits and drawbacks in details. For the constitutive model the material constants will be derived, the model will be developed, and the obtained results are considered for prediction accuracy comparisons. Afterwards, the ANN multi-layer feedforward with backpropagation and Radial Basis Function (RBFN) network will be discussed. These networks can operate as a Blackbox to predict the unknown and highly nonlinear relationship between input and output parameters. Two ANNs feed forward networks with different layers and neurons in each layer, as well as an RBFN are trained and simulated using the MATLAB Toolbox. The RFBN is very efficient for fitting the data especially when there is not a sharp change or interruption in the corresponding true results and there is a function to approximate. The results of all models are compared to each other, first of all in case of a well fitted model by analyzing the statistical measurements such as Correlation Coefficient (R), Average Absolute Relative Error (AARE) and Root Mean Square Error (RMSE). The results demonstrate a clear improvement from the constitutive model to ANN feed forward networks, and later RBFN with 0.999, 2.17 %, and 1.59 for the corresponding modules. The other important criterion is how cumbersome it would be to obtain some models, especially constitutive models constants finding or search of a global minimum for feed forward network. So the RBFN is found to be the best method of training a favorable network. Recently, there are many studies aiming to reduce the number of neurons in the RBFN. Those methods could be really helpful to figure out the important data points giving the most useful information about the stress-strain curves and later finding the optimal experiments giving the perfect results. This could be a great opportunity as reducing the cost of investigation could help to predict behavior of a wide range of materials properly, and pave the way for industries to exploit new desired mechanical properties.