Objectives: Postmortem computed tomography (PMCT) usually includes the generation of great amounts of imaging data, and is often reviewed by forensic pathologists. To allow a more resource-efficient diagnosis, deep neural networks may act as a pre-scanning tool in postmortem radiology. In this study, a deep neural network to classify cases depending on the presence skull fractures on curved maximum intensity projections (CMIP). Methods: Calvarial CMIPs of each 75 cases with and without documented skull fractures were retrospectively generated from our database. Then, half of the data were randomly assigned to either training or validation. In supervised training, fractures were manually marked. During validation, each image received a gradual score between 0 and 1 predicting the likelihood of showing one or more fractures. Results: With a total number of 100 networks trained, the average area under the Receiver Operating Characteristic curve (AUC) was 0.895. The best performing network had an AUC of 0.965. At a classification threshold of 0.79, the network classified fracture cases correctly with a sensitivity of 91.4% and a specificity of 87.5%. Conclusion: Classification based on the existence of skull fractures on CMIPs with deep learning is feasible. For the purpose of pre-scanning PMCT data, a classification threshold of 0.75 with a sensitivity of 100% can be applied. A higher number of images of validated skull fractures available will increase the performance of the network. In the future, Deep learning might enable a more resource-efficient assessment in postmortem radiology.