This thesis investigates the variable production of English /s/ + consonant onset clusters in the speech of 30 adult native Farsi speakers learning English as a second language (L2). In particular, the study examines the development of the homorganic /st/, /sn/ and /sl/ sequences (sC clusters), which are realized variably either via e-epenthesis (e.g., [e[barbelow]st]op) or via its target L2 pronunciation (e.g., [st]op) The sentence reading task as well as the picture-based interview utilized in this investigation followed standard sociolinguistic procedures for data collection and analyses, and included a set of linguistic (e.g., preceding phonological environment, sonority profile of the cluster) and extra-linguistic factors (e.g., level of formality, proficiency in English) whose effects were measured statistically via GoldVarb X.  The results reveal that: (1) the proportion of [e]-epenthesis is higher after a word-final consonant or pause than after a vowel (in which case the sC cluster is resyllabified as two separate syllables, i.e. [V s.C V]); (2) over time (hence with increased L2 proficiency) and in formal situations, the amount of epenthesis decreases, conforming with Major's (2001) Ontogeny Phylogeny Model; and (3) as observed in several studies of L1 acquisition, markedness on continuancy - rather than markedness on sonority - is better able to capture the variable patterns of e-epenthesis in the Farsi-English interlanguage data (i.e., the more marked structures /st/ and /sn/, in which the continuancy feature varies (from [+continuant] /s/ to [-continuant] /t/ and /n/ ) are more likely to trigger the phenomenon of [e]-epenthesis than the less marked nonnative cluster /sl/, in which continuancy is maintained constant (from [+continuant] /s/ to [+continuant] /l/). Based on these results, I analyze the data within a stochastic version of Optimality Theory, and discuss their implications and pedagogical applications for the teaching of pronunciation.