Over the last decade BDD-based symbolic manipulations have been among the most widely used core technologies in the verification domain. To improve their efficiency within the framework of Unbounded Model Checking, we follow some of the most successful trends proposed in this field. We present a very promising approach based on: Mixing forward and backward traversals, dovetailing approximate and exact methods, adopting guided and partitioned searches, efficiently using conjunctive decompositions and generalized cofactor based BDD simplifications. One of the main contributions of this paper is a backward verification procedure based on a prioritized traversal. We call the method "inboundpath-search". Initially, an approximate forward traversal produces overapproximate onion-ring frontier sets. After that, these rings are used as distance estimators and guides to partition state sets in terms of the estimated distance from the "target" set of states. Finally, while the subsequent search is performed, the higher priority is given to the subset with the smallest estimated distance. We experimentally compare our methodology with a state-of-the-art technique (approximate-reachability don't cares model checking) implemented in the freely available VIS tool. Results show interesting improvements in terms of both efficiency and power.