The familial–longitudinal data are collected from a large number of groups or families over a small period of time. This type of data exhibit two-way correlations. First, the responses of the members of a family are likely to be correlated. Second, when the familial responses are repeatedly collected over time, they also exhibit longitudinal correlations. In this paper, we propose a semiparametric linear model with a two-way correlated error structure that accommodates both the longitudinal and familial correlations. We use the generalized least squares method for the estimation of the parametric regression function, whereas the time trend function is estimated by a locally weighted regression smoother. The longitudinal and familial correlations are estimated by using the method of moments. The performance of the estimation methodology is examined through a simulation study by generating the errors of the linear model from multivariate normal as well as heavy-tailed t-distributions. The estimators appear to perform quite well in estimating their parametric and nonparametric population counterparts, including the longitudinal and familial correlations.