Two task were developed to assess second and fourth grade girls' ability to generate quantitative inferences and estimate answers to math word problems. The generation of quantitative inferences is thought to constitute problem representation, and the ability to estimate answers to word problems is facilitated by the presence of an accurate problem representation. It was hypothesized that individual differences in these abilities would be related to mathematics achievement. Twenty second and 20 fourth graders were individually administered simple computation sums and two tasks designed to measure the ability to generate and use quantitative inferences. The Mean Retention Task (MRT) required children to generate a quantitative relational inference and retain it in memory. The Math Inference Task (MIT) consisted of word problems with several alternate answers. The child was required to generate a quantitative inference and use it to choose only those answers that were possibly correct. Fourth graders did better than second graders on computation and the MIT, but there was no difference between grades on the MRT. Performance on the MIT and the MRT was significantly correlated with scores on standardized math achievement tests in fourth grade only. Computation scores were a factor in determining mathematical problem solving ability in second grade only. Finally, MIT and MRT were significantly correlated with each other in both grades. (Author)