Modern Hebrew spirantization (MHS) is a highly variable process with many exceptions. This paper reports the results of an experiment testing the acceptability of variation in alternating and exceptional segments in Modern Hebrew spirantization. Consequently, an Optimality Theoretic (OT) analysis, combining stochastic constraint ranking (Hayes & MacEachern 1998, Boersma 1998, Zuraw 2000, Boersma & Hayes 2001, Hayes 2004) and set-based indexation (Pater 2000) is used to model the results. In MHS the stops [p], [b], and [k] alternate with their fricative counterparts [f], [v], and [χ]. Fricatives occur in post-vocalic position and stops occur elsewhere. This is illustrated in these alternations: [lifgoʃ] ‘to meet’ ~ [pagaʃ] ‘met’ (3p.sg.m.), [livgod] ‘to betray’ ~ [bagad] ‘betrayed’ (3p.sg.m.), and [liχtov] ‘to write’ ~ [katav] ‘wrote’ (3p.sg.m.). Exceptions to MHS are cases in which [p], [b], [k], do not alternate with [f], [v], and [χ] (and vice versa). Rather, the stops may surface in post-vocalic position and the fricatives may surface in non-post-vocalic position, as in these non-alternating paradigms: [likro] ‘to read’ ~ [kara] ‘read’ (3p.sg.m.), and [levater] ‘to give up’ ~ [viter] ‘gave up’ (3p.sg.m.). Variation has been reported in MHS (Adam 2002, Temkin Martínez 2008). This variation consists of segments that normally do conform to the spirantization distribution surfacing as stops where fricatives are expected (as in [fagaʃ] for [pagaʃ]) or as fricatives where stops are predicted (as in [likvor] for [likbor]). The experiment described in this paper examines the nature of variation in both alternating and exceptional segments. Seventy-four native speakers of Hebrew rated the acceptability of the pronunciation of words containing the segments in question. The 204 tokens consisted of words containing the expected and variant forms of both alternating and exceptional segments in word-initial, post-consonantal, and post-vocalic position (embedded in carrier sentences). A statistical analysis of the results shows that variation is acceptable not only in alternating segments, but in exceptional segments as well. However, variation is significantly more acceptable in alternating segments than exceptional ones. This suggests that, despite the relatively high occurrence of variation, speakers still distinguish between alternating and exceptional segments categorically. The OT analysis based on the experimental results combines stochastic constraint rankings to account for variation, and set-based indexation to account for exceptionality. The combined model presented in this paper accounts for the experimental data, allowing for both variation and exceptionality within a single phenomenon across all participants. Additionally, a preliminary analysis of within-subject variation demonstrates that the combined model can further account for individuals’ grammars across word position and segment type.