This paper reports a production experiment examining variation in Modern Hebrew spirantization. Modern Hebrew spirantization is characterized by the alternation of the stops [p], [b], and [k] with their fricative counterparts [f], [v], and [χ], respectively. Typically, fricatives occur post-vocalically, and stops elsewhere, as in (1). (1) Root Infinitive Uninflected Gloss /p/~[f] /pgʃ/ [lifgoʃ] [pagaʃ] ‘to meet’ /b/~[v] /bgd/ [livgod] [bagad] ‘to betray’ /k/~[χ] /ktb/ [liχtov] [katav] ‘to write’ Due to historical mergers and recent borrowings, there are segments that are acoustically identical to those in (1) but that do not alternate, thus potentially forming exceptions (i.e. post-vocalic stops or word-initial fricatives). Exceptions to the labial stop~fricative pairs are non-alternating fricatives ([f] in recent borrowings and [v] from [w] in older forms of Hebrew), while the velars have both a non-alternating stop and non-alternating fricative ([k] from [q], and [χ] from [ħ]). In addition to these exceptions, the alternating segments exhibit variation (Adam 2002, Temkin Martínez, 2010). As with the exceptions, this variation manifests itself as stops and fricatives in positions that are unexpected given spirantization, but in segments that normally alternate. In a perception experiment, Temkin Martínez (2010) found that variation in alternating segments is most acceptable postconsonantally, placing fricatives where stops are expected. Post-vocalically and word-initially, although variation is acceptable, there is a stronger preference for expected segments (fricatives and stops, respectively). The goal of the present study is to determine whether, given ambiguous input, speakers tend to pronounce segments as predicted by spirantization or to maintain the continuancy of the input (treating the target segment as exceptional), and to compare the findings with the perception results in Temkin Martínez (2010). Forty-eight native speakers of Hebrew ages 22 to 46 participated in a sentence-completion task. Verbs from 27 nonce roots were presented auditorily in sentence pairs. Each pair contained a verb in the first sentence, but had it missing from the second sentence. The verbs were inflected so that the target segment’s position would be different in the stimulus and the response, so participants’ realization of each target segment would show whether they perceived it segment as alternating or exceptional. Participants were instructed to complete the second sentence by inflecting the verb from the first sentence for the correct tense. A example sentence appears in (2). (2) [dani ohev lelafem dvarim. Amru li ʃeʔetmol hu_______] Danny loves to NONCE things. Told to me that yesterday he_______ ‘Danny loves to NONCE things. I’ve been told that yesterday he______’ Participants’ production of nonce verbs show variation. Post-consonantally, where stops are expected, there is a strong preference for fricatives, which parallels Temkin Martínez (2010). Also consistent with that study, expected segments were preferred in word-initial and post-vocalic labials. In the velars, however, participants favored non-alternation, producing significantly more fricatives than stops word-initially and more stops than fricatives post-vocalically. This preference for non-alternation in the velars suggests that the existence of both an exceptional velar stop and an exceptional velar fricative gives speakers greater flexibility to produce non-alternating velars than non-alternating labials.