Model checking automatically verifies that a model, e.g., a Labelled Transition System (LTS), obtained from higher-level specification languages satisfies a given temporal property. When the model violates the property, the model checker returns a counterexample, but this counterexample does not precisely identify the source of the bug. Moreover, manually correcting the given specification or model can be a painful and complicated task. In this paper, we propose some techniques for computing patches that can correct an erroneous specification violating an eventually property. These techniques first extract from the whole behavioural model the part which does not satisfy the given property. In a second step, this erroneous part is analysed using several algorithms in order to compute the minimal number of patches in the specification so as to make it satisfy the given property. The approach is fully automated using a tool we implemented and applied on a series of examples for validation purposes.