This paper focuses on quantifying the efficiency of different methods used to allocate resources in flexgrid optical networks. These methods are based on a recently proposed integer linear programming formulation of the routing and spectrum assignment (RSA) problem that takes into account all possible paths and thus theoretically yields optimal solutions, whatever be the objective function. The paper advocates using a metric-based approach for assessing RSA methods preferably to the classical approach based on the blocking probability of dynamic demands because of the long lifetime of optical paths and of the necessity of evaluating an operational network’s state early enough before congestion. The main existing fragmentation metrics are extended to the network level, and a family of network remaining capacity metrics, more suited to assess congestion levels, is introduced. When demands are incrementally mapped, the latter decrease quite linearly, with a slope reflecting the quality of the RSA method. Remaining capacity values are used to compare several off-line methods where the demand sets are mapped either globally or one by one with a suitable ordering. In both on- and off-line cases, using the sum of the demands’ maximum spectrum slice index (an original objective function proposed here), provides the best performance. Finally a method to anticipate a possible congestion, based on a combination of metrics computed on the actual and a reference network instance, is presented.