Network topology discovery is useful for constructing estimates of internal link connectivity from edge measurements. Without any cooperation from the internal routers, topology estimation can be formulated as hierarchical clustering of the leaf nodes based on pair-wise correlations as similarity metrics. We investigate three types of similarity metrics: queueing delay measured by sandwich probes, delay variance measured by packet pairs, and loss rate measured also by packet pairs. Unlike previous work which first assumes the network topology is a binary tree and then tries to generalize to a non-binary tree, we provide a framework which directly deals with general logical tree topologies. Based on our proposed finite mixture model for the set of similarity measurements we develop a penalized hierarchical topology likelihood that leads to a natural clustering of the leaf nodes level by level. A hierarchical algorithm to estimate the topology is developed in a similar manner by finding the best partitions of the leaf nodes. Our simulations show that the algorithm is more robust than binary-tree based methods. The three types of similarity metrics are also evaluated under various network load conditions using ns-2. For background on network topology discovery see references [2-4]. For details on our method see [1].