As widely reported in the scientific literature, audits, benchmarking, safety performance indicators and accident data help the management to understand the current safety performance status and to individuate strong and weak areas of the safety management system. Additionally, being process safety incidents relatively rare, as evidenced in the Baker report on BP Texas City accident, safety performance cannot be measured effectively alone on the basis of such high profile incidents. In the first phase of the research program, process and occupational injuries were studied, collecting field data in a large process industry, over five-year observation. Technical and management improvements seem no longer sufficient to promote safety as at-risk behaviour and unsafe attitudes are still present in spite of all training, supervision and guidance. A thorough analysis on underlying causes connected with human failure was subsequently performed by designing a structured questionnaire, for both in-house and outsourced frontline workers. Data statistical analysis allowed quantifying four conceptual key dimensions within the firm, namely: individual behaviour, organizational climate, human resource management and plants/technology. Significant results were utilized to evidence individual and corporate elements affecting accident frequency for the two workforce types. Conclusions were focused on identifying technical and managerial options to reduce the likelihood of errors and increase risk resilience.