A method is proposed for simultaneous estimation of genetical variance and covariance components for quantitative characters and hence correlations and cross-correlations between relatives. It proceeds by successive approximations, which are not difficult to compute. Judging by examples, they appear to converge rapidly in practice. When distributions are normal and homoscedastic, the method gives the maximum-likelihood estimates and their error variances and covariances. It can be readily modified to take into account assumptions such as that means and variances are equal in fathers and sons. But the amount of calculations required goes up very rapidly with the number of characters and relatives considered stimultaneously.