This study probes the effects of allomorphy on access to Arabic roots and word patterns in two cross-modal priming experiments. Experiment 1 used strong roots which undergo no allomorphy, and weak roots which undergo allomorphy and surface with only two of their three consonants in some derivations. Word pairs sharing a root morpheme prime each other reliably not only when the root was strong (e.g., "participant"/"participate"), but also when it was weak (e.g., "agreement"-"agree," where the weak root [wfq] surfaces fully in the target but not the prime). This facilitation occurred even when the weak root surfaced with different semantic meanings across prime and target (e.g., "destination"/"confront"). Experiment 2 assessed the effects of allomorphy on word pattern processing, comparing word pairs where the word pattern is transparently realised in both prime and target (e.g., "spread"/"bear"], with pairs which share the same underlying word pattern but where a weak root triggers an assimilation process in the prime (e.g., "unite"/"smile"). This assimilation process does not disrupt the CV-structure of the word pattern, in contrast to a third condition where this is disrupted in both prime and target (e.g., [daara]/[qaala] "turn around"/"say"). Strong priming effects were observed in the first two conditions but not in the third. The bearing of these findings on models of lexical processing and representation is discussed.