In this paper I present extensive quantitative evidence showing that it is possible to distinguish different Middle English varieties on the basis of the treatment of word-initial h-, and that it is necessary to postulate that, in some varieties, word-initial h- fails to surface in given contexts, though being present as a consonantal phoneme in the underlying representation. This casts a new light on the old problem of whether word-initial h- was lost in Middle English and restored at a later stage: the data presented here suggest that h- loss was never generalized, though h-less forms did surface as contextual variants of h-ful forms.