A mother tracked her preschooler's number word development daily from 18 to 49 months of age. Naturalistic observations were supplemented with observations during structured (Kumon) training and microgenetic testing. The boy's everyday use of "two" did not become highly reliable and selective for 10 months (at 28 months), emerged later than that of words representing less abstract concepts, and was used in a relatively abstract manner to describe various visible pairs of items. He quickly generalized "two" to partially visible collections and then those that were not visible. Highly reliable use of "one" and "two" appeared to develop simultaneously, before he started using a plural rule, and before he could put out two items upon request. Reliable and accurate use of number words in everyday situations, particularly child-initiated efforts, preceded such use in the contexts of the Kumon training and microgenetic testing, both of which involved adult-initiated tasks. Educational implications include underscoring differences among the first number words by contrasting, for instance, one with two, and pointing out non-examples of a number ("not two") as well as a wide variety of examples, such as "two blocks, two hands, two socks, two airplanes." (Contains 2 figures, 1 table, and 4 footnotes.)