Differing methods of including poverty in analyses of development may produce differing analytic outcomes. Poverty is modeled in a nationally representative sample of maltreated infants using latent curve models, controls for demographic and maltreatment characteristics, and using infant development as the outcome of interest. Poverty was specified as a dichotomous variable (in poverty or not in poverty), a continuous variable (income-to-needs ratio), as low socioeconomic status (SES), as a moderator of developmental predictors, and as being mediated by development predictors for each of four developmental outcomes. Multiple imputation is used to address the problem of missing data. Findings for the complete sample support the use of income-to-needs ratios as the preferred method of measuring poverty based on component and global fit of the model, though effects were generally only found on the intercept factor. The slope factor had few or no predictors, perhaps as a result of relatively small amounts of developmental change in the infants. Some support was found for the more informative mediated and moderated models of poverty as well, and may be of use in the development of interventions to remediate the effects of poverty. Subsamples were created based on gender, membership in a racial minority group, maltreatment type experienced, and type of child welfare placement. In these models, predictors varied compared to each other and the complete sample. Females in foster care and membership in a racial minority were associated with lower scores on the intercept and negative slopes, respectively. When latent dependent variables were used in the latent curve models, fit and precision of estimates improved while the shape of the trajectories did not change. This was similar to prior research.