Among several approaches aiming at the correctness of systems, model-checking is one technique to formally assess system models regarding their desired/undesired behavioural properties. We aim at model-checking the Circus notation that combines Z, CSP, and Morgan's refinement calculus, based on the Unifying Theories of Programming. In this paper, we experiment with approaches for capturing Circus processes in CSP, and for each approach, we evaluate the impact of our decisions on the state-space explored as well as the time spent for such a checking using FDR. We also experimented with the consequences of model-checking CSP models that capture both state invariants and preconditions of Circus models. Some related work on techniques for model-checking Circus was presented by Freitas [12] where a refinement model checker based on automata theory [19] and the operational semantics of Circus [39] was formalised in Z/Eves [34]. He also prototyped a model checker in Java. Moreover, Nogueira et al. [24] also presented a prototype of a model checker based on the operational semantics of Circus within the Microsoft FORMULA [21] framework. However, they could not provide a formal proof of the soundness of their approach, since FORMULA does not have an available formal semantics. Yet another approach for model-checking Circus was defined by Ye and Woodcock [41], who defined a link from Circus to CSP B with model-checking using ProB [31]. Finally Beg [4] prototyped and investigated an automatic translation that supports a subset of Circus constructs. Since CSP M does not have a notion of variables for state as in Z, Circus or even the B-Method, we have to somehow capture them in order to obtain a CSP M model as similar as possible to the original Circus one. Therefore, one could either use a memory model [30,25] in order to manage the values of the state variables, or else, to adopt the idea of state-variable parametrised processes [4]. Following the results presented in ABZ'16 [16], which involved manual translation, we decided to develop Circus2CSP 3 , an automatic translator from Circus into CSP M , aiming at model-checking with FDR. Our tool was then built based on the strategy presented in Section 5.3 of Deliverable 24.1 [29], from the COMPASS project [37], that defines a rigorous but manual translation strategy aiming at obtaining CSP M specifications from Circus. This paper reports design decisions regarding different approaches for model checking and experimental results obtained for Circus specifications. Such experiments were enough to identify an effective general form for any CSP M model derived from Circus, where FDR could perform refinement checks with reduced time and memory consumption compared to existing approaches from the literature. 2 Circus Background A Circus specification is in some sense an extension of Z [40] in that it takes the paragraphs of Z and adds new paragraph forms that can define Circus channels, processes and actions. Channels correspond to CSP events: channel c : T Circus actions can be c...