Children learn to count, and even learn the cardinal meanings of the first three or four verbal numerals ("one" through "three" or "four"), before they master the numerical significance of counting. If so, it follows that the cardinal meanings of those first few numerals cannot be derived, initially, from their place in the count list and the counting routine. What non-verbal representations, then, support the cardinal meanings of verbal numerals before children have mastered how counting does so? Four experiments addressed the commonly adopted assumption that in the earliest period of learning the meanings of number words, children map verbal numerals to regions of the analog number system (ANS), a system of representation with numerical content that is widely attested in animals and in human infants. Experiment 1 confirmed that children who know what "three" means, but who do not yet know what "four" means, and do not yet know how counting represents number, can be easily taught the meaning of "four," if they are trained to indicate sets of four when they are paired with a series of sets that contrast numerically with four. If children learn "four" by mapping the word to an ANS representation of sets of four, and if such ANS value-to-word mappings underlie the meanings of other known numerals early in development, then analogous teaching should enable young children to establish a ANS value-to-word mapping for between "ten" and sets of 10 as specified by the ANS. Furthermore, the ease of learning should be a function of the ratio of the number of individuals in the comparison set to 10. Three further experiments tested these hypotheses by attempting to teach young Cardinal Principle-knowers the meaning of the word "ten," under the same training conditions "three-"knowers are easily taught the meaning of "four". The children learned which picture in each training pair had "ten." However, test trials with novel animals and spatial configurations showed that they had failed to learn what set sizes should be labeled "ten", even when, after training, they were asked to indicate a set of 10 vs. a set of 20 or 30 (well within the ratio sensitivity of the ANS even early in infancy). Furthermore, there was no effect of ratio on success during test trials. These data provide new evidence that ANS value-to-word mappings do not underlie the meanings of number words early in development. We discuss what other non-verbal representations might do so, and discuss other ways the ANS may support learning how counting represents number.