Since Ross's first formulation of Left Dislocation (LD) as a movement rule (Ross (1967, ch. VI)), a number of people have challenged his analysis, proposing instead that the lefthand constituent be base-generated (Postal (1971); Hirschbuhler (1974; 1975); Rodman (1974); Van Riemsdijk and Zwarts (1974); Gundel (1975)). Before briefly considering some of their arguments and alternative proposals, which I will argue focus on a quite different construction, I will present what I take to be rather strong evidence in favor of a movement analysis, for at least a large class of sentences in at least some languages. All of the arguments I offer below have the following form: (1) There is some rule operating on NPs that ordinarily displays either a governor or a trigger or a controller to the left of the affected NP. (2) In our "LD" data the affected NP appears to the left of such a governor (or trigger, or controller) rather than to its right. (3) Were we not to posit a movement rule that applies (we must assume) after the relevant rule has operated gn the NP, we would be compelled to state the same restrictions twice, as if they were independent ones, thus missing a basic regularity.'