Precise predictive models are required for the use of machine learning methods for quality control in injection molding. Thermal images offer the advantage of containing information in the data that is not available in machine and process data. Currently, convolutional neural networks (CNN) have numerous applications in image recognition. Therefore, the objective of this work was to investigate the application of convolutional neural networks to thermal images of injection molded parts. For this purpose, 751 injection molding cycles from a central composite design were used. The goal was to predict the weight, height, and width of the injection molded part. The results were also compared with classical machine learning methods. Depending on the quality parameters, the networks were able to achieve an R² of up to 0.91 and were thus among the three best methods.