The problem of estimating intraclass correlation is considered under the realistic setting where the number of individuals may vary from group to group. A simple, efficient, iterative method of computing the maximum likelihood estimates of components of variance and intraclass correlation under the assumption of multivariate normality is provided. The method is compared with some existing methods using simulated data and is found to perform very well. The method is also applied to some real life data on human dermotoglyphic characters. Although the application considered is genetic research based on family data the method is equally relevant to other applications, such as epidemiological and educational research.