This dissertation explores heterogeneity in how family structure affects children. Detailed measurements of parental interactions are exploited to estimate a dynamic economic model of parental relationship quality, parental decisions to continue their relationship, and child human capital development. The empirical specification takes into account measurement error, controls for observed demographics and initial conditions, and integrates out missing data. Parental separation is shown to have little effect on cognitive skill development during childhood, but is found to exert considerable influence on the development of non-cognitive skills such as the ability to control aggression. Parental separation's effect on non-cognitive skill development largely depends on parental relationship quality; in particular, children whose parents have a poor quality relationship on average benefit if their parents separate. If separated parents in the sample had instead chosen to stay together, their children would have on average been 21 percent more likely to receive special services because of Attention Deficit Disorder (ADD) and 20 percent more likely to have been suspended or expelled. Almost two-thirds of the difference in non-cognitive skills between children of separated and non-separated parents at age 9 can be explained by differences in parental relationship quality, while only 16 percent can be explained by differences in observable demographics.