A study in an early French partial immersion program in a Melbourne (Australia) elementary school investigated participating children's performance in mathematics in English (their first language) and French. Mathematics testing in both languages has been undertaken regularly as part of a long-term evaluation of the program. In the first year of testing (1995), there were no significant difference in results of students taking the test in English or French. However, in 1996 a difference emerged, with grade 5 students taking the test in English doing significantly better than those taking the test in its French version. Item difficulty analyses were conducted to reveal the problematic questions, and a content analysis was performed on the aberrant items. The study revealed new information about how children process mathematics problems in their second language, and provides insights into the development of the students' language in a partial immersion program. (MSE)