This paper proposes an extension of Tree-DOP which approximates the LFG-DOP model. GF-DOP combines the robustness of the DOP model with some of the linguistic competence of LFG. LFG c-structure trees are augmented with LFG functional information, with the aim of (i) generating more informative parses than Tree-DOP; (ii) improving overall parse ranking by modelling grammatical features; and (iii) avoiding the inconsistent probability models of LFG-DOP. In a number of experiments on the HomeCentre corpus, we report on which (groups of) features most heavily influence parse quality, both positively and negatively.