In this paper I discuss the distribution of the so-called stem marker -ku- in Swahili, which is found with some, mostly monosyllabic verbs when used with certain tense morphemes. After introducing the data, I show that there are three potential analyses to explain the data, which I discuss in turn. Two of these analyses are phonological in nature and relate the distribution of -ku- to stress assignment. The first stress related analysis, which I call the `prevent-stress`rule is popular in textbooks and teaching manuals of Swahili (e.g. Ashton 1944, Russell 1996, Maw 1999), but turns out to be wrong. According to the second, more satisfactory stress rule, first proposed by Meinhof (1910a, 1910b, 1933), -ku- is deleted throughout except in stressed syllables and before vowels. While this analysis provides a sufficient diachronic account of the distribution of -ku-, I provide an alternative synchronic analysis, according to which the distribution of -ku- results from a number of alternative lexical entries for the relevant verbs.