Two three-layer artificial neural network (ANN) models were respectively developed to predict biogas and methane production rates in a pilot-scale mesophilic up-flow anaerobic sludge blanket (UASB) reactor treating molasses wastewater. Eight process-related variables such as volumetric organic loading rate (OLR), influent and effluent pH, operating temperature, influent and effluent alkalinity, effluent chemical oxygen demand (COD), and volatile fatty acid (VFA) concentrations were selected for the implementation of an artificial intelligence-based approach. A tangent sigmoid transfer function (tansig) at the hidden layer and a linear transfer function (purelin) at the output layer were conducted for the proposed ANN models. Several benchmark comparisons were conducted to obtain an optimal algorithm for training the network. After backpropagation training combined with principal component analysis (PCA), the scaled conjugate gradient algorithm (trainscg) was found as the best of the 11 training algorithms. The number of neurons in the hidden layer was optimized as nine and 12 with the minimum mean squared errors (MSE) of 0.06238 and 0.06488, respectively, for the estimation of biogas and methane production rates. ANN-predicted results were also compared to the outputs of two non-linear regression models by means of several statistical indicators, such as determination coefficient (R-2), unsystematic root mean-square error (RMSEU), index of agreement (IA), and fractional variance (FV). Computational results clearly demonstrated that, compared to the conventional multiple regression-based methodology, the proposed ANN-based models produced smaller deviations and exhibited superior predictive accuracy with satisfactory determination coefficients of about 0.935 and 0.924, respectively, for the forecasts of biogas and methane production rates.