During the melting and processing practices, an accurate knowledge of liquidus temperature is necessary in the determination of process parameters relating to a given alloy. Adaptive neuro-fuzzy inference system (ANFIS) modeling method has been used to improve accuracy of prediction for liquidus temperature based on the compositions of Al-Si series cast alloys. The developed fuzzy inference system could extract Takagi-Sugeno type fuzzy rules from data directly, and has a feed-forward network structure with supervised learning capability. In order to adapt the parameters of the model, the proposed fuzzy inference system is trained over a wide range of compositions from the published data of industrial alloys. The result shows that, the developed ANFIS model can capture non-linear relationships between compositions and liquidus temperature, and then provides better prediction than the reported multiple statistic analysis. The developed model can be used to predict the liquidus temperature needed in computer modeling and thermodynamic calculation, which are needed in the aluminium alloys casting industry and research.