Abstract <p>Photovoltaic solar power generation has shown remarkable growth. Understanding the influence of climatic factors and variables on energy production enables the development of predictive models and the assessment of generation potential across different regions. In countries such as Brazil, the availability of surface solar radiation (<i>R</i>_s) data can be limited, making models based on more widely available climatic variables highly relevant. The aim of this study was to use data from a conventional climatological station to develop multiple linear regression (MLR) models that are easy to implement and provide good predictive performance for estimating energy generation in a residential on-grid photovoltaic system located in Minas Gerais, Brazil. Surface radiation was estimated from sunshine duration using the Angström-Prescott model, and several climatic and derived variables were evaluated. MLR models were built and tested for their predictive ability using both daily and monthly energy generation data. Estimated <i>R</i>_s alone explained 81% of the variability in energy generation. The best-performing MLR models, which included <i>R</i>_s, sunshine duration, and Julian day, achieved an <i>R</i>² of 0.854. A model relying solely on the station’s native variables, without the need to calculate <i>R</i>_s, also showed similar performance. Compared with XGBoost, MLR models achieved comparable results while requiring fewer variables, offering greater simplicity and interpretability. The findings demonstrate the applicability of the proposed approach, particularly in regions lacking radiation data, and provide a foundation for the development of predictive models in other regions and photovoltaic systems, while also highlighting the most relevant variables.</p>