<p>An artificial neural network (ANN) was developed to predict the energies of molecular crystals. The model is trained with simple alcohol crystalline structures and it is shown that it correctly predicts energy difference between crystalline structures of longer simple alcohol crystalline structure. The method can be extended to other molecular crystalline systems. The ANN is based on the Behler–Parrinello neural network framework originally developed for molecular systems, in which the total crystal energy is expressed as a sum of local atomic energy contributions. To adapt the method to crystalline environments, a 3 × 3 molecular supercell was constructed to account for all nearest neighbors surrounding each atom and to generate accurate descriptors of the local atomic environments. Element-specific feedforward neural networks were employed to evaluate the energy contribution associated with each atomic environment. The accuracy and transferability of the model were assessed using multiple test sets composed of molecular crystals not included in the training data. The predicted energies show good agreement with density functional theory (DFT) calculations. However, the current description of atomic environments leads to high computational costs for dense crystalline systems, indicating the need for more efficient descriptors.</p>