Children in the second-grade classrooms of three rural schools (n 5 150) completed a variety of psychometric and curriculum-based tests, and were rated by their teachers and parents on dimensions of their everyday behaviour; demographic data (e.g. socioeconomic status, presence of mother in the home) and biographical information (e.g. gender, age, birth order) were also collected for each child. Some of these data (e.g. child's age and gender) were more cost-ef cient to collect than others (e.g. parent ratings). Measures were evaluated in terms of their salience for constructing a multivariate model that would predict subsequent grade 2 outcome, with the most cost-effectiv e variables being inserted rst. In this way, both the cost-ef ciency and predictive power of independent variables (IVs) were taken into consideration when attempting to build a predictive model. A model containing three IVs (scores on curriculum-based tests, teacher ratings of children's attention span, and teacher ratings of helpfulness) ultimately predicted 51% of the variance in grade 2 outcome. These results demonstrate, rst, that it is possible to build a relatively strong predictive model of grade 2 outcome, although not based on variables that are cheap and quick to measure. Second, that doing well in grade 2 is not so much a matter of having well-develop ed, broad-ranging psychometric abilities, but more a matter of mastering elements of the at OhioLink