This paper deals with the use of Russian numerals in spontaneous speech of Moksha-Russian and Hill Mari-Russian bilinguals. Based on a comparative corpus study of numeral phrases with code-switching in Moksha and Hill Mari, we propose an analysis in terms of the Matrix Language Frame (MLF) model. We discuss the factors that might influence the choice of language: the numerical value, the type of the context, and the syntactic type of the numeral. Although the numerical systems are maintained in each of these Uralic languages, there is a strong tendency to use Russian numerals to express larger quantities. We argue that ordinal numerals are switched more frequently for structural reasons: they do not occupy the same position as cardinal numerals in the noun phrase. We also argue that the formation of embedded language islands is influenced by the types of dependencies established in a construction (quantifier-like cardinal numerals vs. adjective-like ordinal numerals), which supports Muysken’s model of categorical equivalence as a condition for code-switching.