As semi-solid forging (SSF) is compared with conventional casting such as gravity die-casting and squeeze casting, the product without inner defects can be obtained from semi-solid forming and globular microstructure as well. Generally, SSF consists of reheating, forging, and ejecting processes. In the reheating process, the materials are heated up to the temperature between the solidus and liquidus line at which the materials exists in the form of liquid-solid mixture. The process variables such as reheating time, reheating temperature, reheating holding time, and induction heating power has large effect on the quality of the reheated billets. It is difficult to consider all the variables at the same time for predicting the quality. In this paper, Taguchi method, regression analysis and neural network were applied to analyze the relationship between processing conditions and solid fraction. A356 alloy was used for the present study, and the learning data were extracted from the reheating experiments. Results by neural network were in good agreement with those by experiment. Polynominal regression analysis was formulated using the test data from neural network. Optimum processing condition was calculated to minimize the grain size and solid fraction standard deviation or to maximize the specimen temperature average. Discussion is given about reheating process of row material and results are presented with regard to accurate process variables fur proper solid fraction, specimen temperature and grain size.