It is generally admitted that disyllabic words in English are stressed according to their morphological make-up. While prefixed words show differential behaviour according to major grammatical category, non-derived nouns are allegedly trochaic and underived verbs are either iambic or trochaic following rules of quantity-sensitivity. This paper presents a database which was compiled in order to test native speakers' intuition about the stress of disyllables. 53 nonsense words were created displaying different phonological and morphological structures forced by the spelling. These words were embedded in sentences so that each form appears twice, once as a nominal and once as a verbal form. We recorded 20 speakers reading 106 sentences giving 2120 tokens. The construction of nonce words is the main issue at stake, the paper is therefore concerned with methodological questions regarding the design of a heuristic corpus. The data will be freely available on SLDR for the scientific community.