Word segmentation is an important step in NLP preprocessing, even more so in languages such as Modern Standard Arabic, where script conventions amalgamate words and whose morphology is complex. This paper focuses on grammatical segmentation of words, clitics and inflectional affixes. Among the segmenters proposed in the literature and recorded here, we selected twelve segmenters to be evaluated on linguistic and grammatical accuracy instead of information retrieval, on two tasks, detection of the stem and segmentation of clitics and affixes. We present the gold standard corpus we devised to make this evaluation. Results show that Farasa has an best overall performance; Madamira is able to disambiguate in some cases and is better in particles, Stanford segmenter is third. But all three of them lack efficiency to correctly segment verb forms, showing the need for better tools. © 2025 IEEE.