We study the following generalization of the classical edge coloring problem: Given a weighted graph, find a partition of its edges into matchings (colors), each one of weight equal to the maximum weight of its edges, so that the total weight of the partition is minimized. We present new approximation algorithms for several variants of the problem with respect to the class of the underlying graph. In particular, we deal with variants which either are known to be NP-hard (general and bipartite graphs) or are proven to be NP-hard in this paper (complete graphs with bi-valued edge weights) or their complexity question still remains open (trees).