Breadth-first strategy is described as a basic concept of molecular path generation. Minor modifications on the algorithm also make depth-first strategy available. Breadth-first strategy is advantageous in topological distance type problems. It is applied to generate path counts as topological descriptors. Complete path count sets are compared with atom type, bond type and sphere count descriptor sets by principal components analysis. The tendency of degeneration and the orthogonality of the descriptor sets are investigated. Pure descriptor set models and mixed models are developed to predict boiling points and retention index data of alkanes. It is shown that a mixed model incorporating path count and bond type descriptors covers nearly the whole topological information of the investigated compounds. The estimation error of this model lies near an approximated pure experimental error.