This paper presents a computer aided model-based test generation method. We propose this approach as a complement to the LTG (Leirios Test Generator) method, which extracts functional tests out of a formal behavioral model M by means of static (or structural) selection criteria. Our method computes additional tests by applying dynamic (or behavioral) selection criteria (test purposes called TP). Applying TP directly to M is usually not possible for industrial applications due to the huge (possibly infinite) size of their state space. We compute an abstraction A of M by predicate abstraction. We propose a method to define a set of abstraction predicates from information of TP. We generate symbolic tests from A by using TP as a dynamic selection criterion. Then we instantiate them on M, which allows us play the tests on the implementation the same way as we play the functional ones. Our experimental results show that our tests are complementary to the structural ones.