The construction industry is one of the most dangerous industries worldwide, with many workers involved in accidents annually. The main cause of accidents is that hazards are not recognised, making safety training vital. Virtual reality (VR) technology offers a new approach to construction safety training. Therefore, exploring the hazard identification performance of individuals with different learning styles is what highlights our innovation in matching different training styles to different individuals. This study evaluates the performance of individuals using two training modes. Visual search eye-movement indicators were collected during hazard identification to explore their effects on hazard recognition performance. Simultaneously, learning style theory was introduced to classify individuals using eye-tracking technology and then investigate the differences in hazard identification performance enhancement of individuals with different learning styles after different safety training. The results showed that VR safety training improves individual hazard identification performance more than traditional safety training, but VR safety training is not necessarily suitable for everyone. Reflective, verbal, and global learners can achieve greater improvement by participating in traditional safety training; active, intuitive, and sequential learners can achieve greater improvement through VR safety training; and sensing and visual learners do not exhibit a significant difference in improvement by participating in the two training modalities. Therefore, sensing and visual learners can choose to participating in traditional safety training, which is less costly and easier. This study provides meaningful recommendations for personalised construction safety training.