The work covers the issues related to the diagnosis of selected occupational safety hazards in Polish wood processing enterprises. The main objective is to build models in order to identify occupational accident profiles in these enterprises on the basis of individual records characterizing the casualties, provided by Statistics Poland. The modelling task employed the latent class analysis (LCA) data mining technique. In order to enhance the process of building the LCA model and to support the procedure of selecting input variables relevant to the model, an iterative algorithm was elaborated by the authors. The impact of an enterprise size on occupational accident consequences was statistically confirmed. Following this result, LCA models were developed independently for smaller (micro and small), and for larger (medium and large) enterprises. Latent classes, presenting occupational accident profiles, were visualized in the form of heat maps. Similarities and differences between the occupational accident profiles identified for the two types of enterprises were indicated. It has been shown that employees of smaller enterprises are at greater risk of suffering more serious injury from accidents at work than employees of larger enterprises. However, in both cases, the most critical latent classes concern occupational accidents related to operating machinery; they affect workers with a low level of job seniority, and result in injuries (often traumatic amputations) involving upper limbs in particular.