Methods to account for population structure (PS) in genome-wide association studies have been well developed in samples of unrelated individuals, but when a sample is composed of families, the task of finding and accounting for PS is not as straight forward. Family-based tests that condition on parental genotypes or their sufficient statistics are immune to biases due to PS, but are known to have low power, particularly for unselected samples. Population-based approaches that use all available data are an attractive alternative, but the methods have not been evaluated for continuous outcomes when a sample has both family and PS. Therefore, we compare through simulation the performance of population-based regression models that account for family and PS with continuous outcomes using a range of family sizes and structures, including two and three generational families with admixed and discrete PS. We find that when computation time is a concern, the Dupuis et al. efficient score test performs very well. When computational time is not an issue, a linear mixed effects model adjusting for genetic principal components tends to have slightly better power than the score test and may be preferred.