How children acquire the rules governing past tense production has been for many years a test bed case for nativist-constructivist debates about the nature of innate knowledge and its role in language acquisition. However, previous studies have tested the acquisition of past tense via corpus analysis, in which errors are rare, or elicitation tasks, in which tested items are few, resulting in limited between-item variability. To address these weaknesses, we analysed data from a uniquely large and longitudinal dataset containing 694 verbs, collected via an educational online platform. We examined whether form-frequency, phonological neighbourhood density (PND), and telicity predict the verb-level difficulty of past tense forms in Dutch. Our sample consists of Dutch-speaking children aged 8-12 years old, the age at which children are still making past tense over-regularisation errors. Analyses are ongoing but preliminary results suggest a role for all three factors and an interaction between frequency and PND.