Fiber composites must be evaluated to achieve correct use in various fields. Their properties, performance, condition, and integrity can be quickly predicted and optimized by machine learning (ML), after extensive training, compared with experiments and conventional computational simulations. In this document, papers on ML applications in fiber composites were collected and critically reviewed. It was revealed that kind learning environments have been primarily used. Supervised ML has been more frequently used than unsupervised ML, whereas some specific semi–supervised ML (e.g., reinforcement learning) or deep predictive control have been overlooked. Most ML applications have been successful on the laboratory scale and in the short term. Furthermore, the deployment of ML applications has been overlooked. In addition, retroactive feedback from the manufacturing of fiber and polymers to the manufacturing of composite laminates and structures was neglected. Accordingly, a control loop in the chain of manufacturing processes was discussed. Additionally, language processing tools and statistics were used to summarize and analyze the papers. Finally, it was proposed that multiscale modeling using ML and physics is a potential approach to advance predictions for future applications. Therefore, physicochemical interactions (van der Waals or electrostatic) from nanoscale can be included.