This paper presents two experimental studies that examine English-speaking learners' knowledge of unaccusativity in Japanese. Following Levin & Rappaport Hovav (1995) and Kageyama (1993), I distinguish between deep versus surface unaccusativity. Deep unaccusativity refers to a representation where the argument of an unaccusative verb is base-generated in the object position, whereas surface unaccusativity refers to a configuration where the unaccusative argument remains in the object position even at a later stage of derivation (i.e., it does not move to the subject position). It has been claimed that Japanese exhibits both deep and surface unaccusativity (Kageyama, 1993; Yatsushiro, 1999). Previous findings in second language (L2) literature suggest that learners of Japanese are sensitive to deep unaccusativity, based on the results of tests involving the takusan 'a lot' construction and quantifier floating (Hirakawa, 1999; Sorace Shomura, 2001). However, these studies failed to tap knowledge of surface unaccusativity where Case Drop was used as a diagnostic. The failure was due to the fact that even native speakers did not behave as the theory predicted. Following up on these studies, the present paper uses new constructions in order to examine the issue further : resultatives for deep unaccusativity and causative-passives for surface unaccusativity. Results confirm that deep unaccusativity is observed by both intermediate and advanced learners but that surface unaccusativity is observed only by advanced learners. I argue that L2 learners' grammars are constrained by the principles of UG but that the L1 also has some effects in the acquisition of the properties of unaccusative verbs in the L2 when they differ from those in the L1.