Considering Sentiment Analysis’s importance, the Natural Language Processing (NLP) field has seen a surge in studies dedicated to this task. However, research efforts in Arabic Natural Language Processing haven’t reached the same prominence as those in non-Latin alphabet languages. This discrepancy can be attributed to the specificity of the Arabic language and the limited availability of freely accessible lexical resources. Considering these challenges, our paper focuses on Sentiment Analysis in Modern Standard Arabic. We achieve this using pre-trained Arabic BERT models, specifically AraBERT, ALBERT, CAMeLBERT, AraELECTRA, and QARIB. Our approach was tested on the 100k reviews dataset and Arabic Moroccan Arabic corpus (MAC) dataset. Notably, with the MAC dataset, our proposed system at- attains an accuracy of 96% when using the QARIB model, 93% when using Ara- BERT and 26% with CAMeLBERT. Regarding the 100k reviews dataset, the best accuracy of 93% was achieved when using AraBERT, followed by QARIB with an accuracy of 89% and lower accuracy was when using ALBERT with 39%. We compared the research results obtained through our approach with those of other leading methods, demonstrating the effectiveness of our methodology. These findings provide valuable insights for future enhancements in this field.