The coding guidelines were developed in the context of the ANR-FNS funded project REM – ‘Re-thinking English Modal constructions: From feature-based paradigms to usage-based probabilistic representations’ (https://anr.fr/Project-ANR-16-CE93-0009), coordinated by Ilse Depraetere (ULille) and Martin Hilpert (Université de Neuchâtel). One of the main aims of the project was to get a clearer view on the factors (lexical, formal, semantic, and pragmatic) that enable English speakers to choose one among the following modal expressions: can, could, may, might, be able to – on the possibility side – and should, must, have to, need to and ought to – on the necessity side. For this study, 36 predictors were identified and defined. The overview and description that follow are meant to help the annotator analyse English modal sentences in terms of these predictors. The REM dataset, which contains 5,000 instances extracted from COCA, each of which was coded in terms of these features, is available from the companion website of the book Models of Modals.