Studies testing the knowledge of syntactic properties have resulted in two potentially contrasting proposals in relation to third language acquisition (TLA); the Cumulative Enhancement Model (Flynn et al., 2004), which proposes that previously learned languages will positively affect the acquisition of a third language (L3); and the "second language (L2) status factor" hypothesis (Bardel and Falk, 2007), which proposes that the primacy of the L2 can block the potential positive effects that may be transferable from the first language (L1). This article attempts to extend these hypotheses to the domain of morphosyntax, in relation to the TLA of the properties of grammatical number and gender concord marking on German attributive adjectives; these properties not present in the L1 of Japanese, or the L2 of English. Two further factors are of interest in the current study; first, the performance of the learners according to their L3 and their L2 proficiency levels, a variable not discussed in the above-mentioned studies; and, second, the role that the type of task has on the performance of these learners. Three groups of Japanese native speakers (matched for proficiency within each German group), but with differing English proficiencies, completed a carefully balanced gap-filling task, together with two oral elicitation tasks in the form of games; both of these elicited tokens of adjectival inflection. Initial results offer partial support for weaker versions of the two hypotheses mentioned above. However, neither of the L3 models tested can fully account for the results obtained, which are more consistent with a feature-based account of the organization of grammar in the domain of morphosyntax, such as that of Distributed Morphology (DM) (Halle and Marantz, 1993). DM is a model for language acquisition which--coupled with a view that the Subset Principle proposed by this account is not observed by non-primary language learners--has recently been proposed to explain the optionality observed in L2 learners' production (Hawkins et al., 2006). The data presented here suggest that it could be extended to L3 learners' production. (Contains 16 tables, 3 figures and 14 notes.)