The present paper is concerned with a new statistical approach, involving latent and manifest variables applied in order to assess the corporate social responsibility performance for small and medium-sized industrial enterprises (SMI). To achieve this aim, our approach consists first to identify five significant components that are suitable to build a Social Responsibility Performance Index Model adapted to the SMI specifications: Governance and Management, Human Resources, Safety and Health at Work, Environment and Sustainable Development, Business Relationship and Best Practices. Then the structural equation model (SEM) is stated to describe various causal connections between latent and observed variables. The SEM’s resolution is based on the Partial Least Squares (PLS) method and the implementation is running in the XLSTAT software by exploring the data collected from a survey questionnaire intended for the different stakeholders concerned with the SMI activities. The obtained outputs could be examined and analysed in order to develop plans and actions for improvement in the CSR.