Fluidized bed melt granulation has recently been recognized as a promising technique with numerous advantages over conventional granulation techniques. The aim of this study was to evaluate the possibility of using response surface methodology and artificial neural networks for optimizing in situ fluidized bed melt granulation and to compare them with regard to modeling ability and predictability. The experiments were organized in line with the Box–Behnken design. The influence of binder content, binder particle size, and granulation time on granule properties was evaluated. In addition to the response surface analysis, a multilayer perceptron neural network was applied for data modeling. It was found that in situ fluidized bed melt granulation can be used for production of spherical granules with good flowability. Binder particle size had the most pronounced influence on granule size and shape, suggesting the importance of this parameter in achieving desired granule properties. It was found that binder content can be a critical factor for the width of granule size distribution and yield when immersion and layering is the dominant agglomeration mechanism. The results obtained indicate that both in silico techniques can be useful tools in defining the design space and optimization of in situ fluidized bed melt granulation.