This paper aims to present an expert system for fault detection in refrigeration systems. Four faults were considered: condenser fouling, evaporator freezing, charge leakage and infiltration of non-condensable gases. For the successful application in real refrigerators the fault detection algorithm must use a minimal instrumentation set and must not depend on a large set of configuration tests. Therefore, the expert system proposed in this paper needs only three extra temperatures measurement and electrical variables already available from compressor motor driver. Another constraint considered was to avoid false positives as door openings and defrost events. The algorithm for fault detection was experimentally developed for a light commercial refrigeration system used to store ice cream but could be applied for a wide range of refrigeration systems. The expert algorithm was validated with experimental data at three ambient temperatures (20°C, 25°C and 32°C) and reached 92.6% of accuracy. Algorithms based on artificial intelligence were trained with data obtained at 25°C and 32°C and further compared to expert algorithm at 20°C.