The 'prediction approach' proposed by Dearden, Machin and Reed (DMR) consists in (1) regressing the observed incomes of the child and parent families on separate sets of predetermined variables, and (2) regressing the child's predicted income on that of the parents. Conceptually, this estimator must relate to the 2SLS/IV estimator. We re-derive the prediction estimator in matrix form, and reconsider its consistency requirements. The measurement model of DMR is then embedded within a simultaneous equations framework, for which an alternative 2SLS/IV estimator is proposed. The latter produces larger estimates for the intergenerational correlation. The policy relevance of the two sets of findings is then discussed.