In a recent paper \textit{T. Amemiya} [ibid. 50, 689-711 (1982; Zbl 0493.62098)] introduced a class of estimators, termed two-stage least absolute deviations (2SLAD) estimators, for the parameters of a structural equation in a simultaneous equations model; the estimators are defined to minimize a sum of absolute values of residual terms which depend upon preliminary (''first-stage'') estimates of reduced form parameters. However, asymptotic normality of the 2SLAD estimators was proved only for normally-distributed error terms, for which more efficient estimation methods (such as 2SLS) would be preferred. In this study asymptotic normality of the proposed estimators is demonstrated for more general distributions of the error terms, thereby strengthening the conclusions to be drawn from Amemiya's efficiency comparisons. In addition, expressions for the large-sample covariance matrices are given which are somewhat simpler than those in Amemiya's paper. Finally, an asymptotic equivalence proposition is shown for the sub-class of ''double'' two-stage least absolute deviations estimators; this result is analogous to the finite sample equivalence of \textit{H. Theil's} interpretation of 2SLS [Repeated least squares applied to complete equation systems, Mimeo, Central Planning Bureau, The Hague (1953)] and its instrumental variables interpretation.