I present a filter method for solving general nonlinear and nonconvex constrained optimization problems. The method is of the filter variety, but utilizes an exact penalty function for computing a trial step. Particular advantages of our approach include that (i) all subproblems are feasible and regularized; (ii) a typical restoration phase that focuses exclusively on improving feasibility is replaced by a penalty mode that accounts for the objective function and feasibility; and (iii) a single uniform step computation is used during every iteration of the algorithm. An additional contribution is a strategy for defining the so-called filter margin based on local feasibility estimates available as a byproduct of our step computation procedure.