Unlike other animal species, humans have the ability to represent large exact quantities. While different theories in number cognition have attributed this ability to our access to natural language, the question of how exactly natural language affords humans this unique ability remains unclear. Studying bilinguals provides a valuable approach to investigating the relationship between numbers and language, as documenting the similarities and differences across languages can inform us about the role of language-specific experiences in the development of numerical representations. In this dissertation, I will argue that language- specific experiences play a fundamental role in the earliest steps of number acquisition, drawing on the evidence from 3- to 7-year-old bilingual children. In Chapter 1, I present evidence that French-English bilinguals estimate the numerosity of arrays of dots differently across their two languages. This asymmetry in bilinguals’ mappings between number words and non-verbal representations across their two languages reveals that these mappings rely on language- specific knowledge of the structure of their count lists. In Chapter 2, I investigate some potential methodological issues when testing bilingual children and show that bilingual studies should take into account how test-retest reliability can contribute to observed differences across languages in bilinguals. However, in the case of Give-a-Number, some levels are more affected than others. Finally, in Chapter 3, I explore further the role of language-specific experiences in the mappings between number words and non-verbal representations by showing that children who know how to count do not subitize similarly across their two languages. Instead, differences in subitizing skills across languages suggest that language-specific experiences play a role from the very beginning of number word acquisition. Together, these studies suggest that language-specific experiences play a major role in building mappings between number words and non-verbal representations, via the estimation of large and small sets. These studies also reveal that some basic numerical abilities don’t transfer across languages in bilinguals.