We present a generic study of unambiguous discrimination between two mixed quantum states. We derive operational optimality conditions and show that the optimal measurements can be classified according to their rank. In Hilbert space dimensions smaller or equal to five this leads to the complete optimal solution. We demonstrate our method with a physical example, namely the unambiguous comparison of n quantum states, and find the optimal success probability. PACS numbers: 03.67.-a,03.65.TaAccording to the laws of quantum mechanics, two nonorthogonal quantum states cannot be distinguished perfectly. This fact has far-reaching consequences in quantum information processing, e.g. it allows to generate a secret random key in quantum cryptography. In spite of the fundamental nature of the problem of state discrimination, determining the optimal measurement to distinguish two (mixed) quantum states is far from being trivial.In the literature, two main paths to state discrimination have been taken [1]: firstly, in minimum error discrimination, the unavoidable error in distinguishing two states from each other is minimized. This problem has been completely solved in Ref. [2]. Secondly, in unambiguous state discrimination (USD), no error is allowed, but an inconclusive answer may occur. The optimal USD measurement minimizes the probability of an inconclusive answer [3][4][5]. Although USD has found much attention in the recent years, and special examples have been solved, no general solution is known so far for the case of mixed states. A strategy that is analogous to USD, but applicable also for linearly dependent states are maximum confidence measurements, discussed in Ref. [6].The aim of the current contribution is to present the optimal USD measurement for cases which cannot be reduced to the pure state case and thus to known solutions. This analysis can be applied for the unambiguous discrimination of any two density operators acting on a Hilbert space up to dimensions five. This goes beyond previous results which require a high symmetry or other very special properties of the given states [7][8][9][10][11][12][13]. We will show the main ideas and steps towards the solution, and explain the technical details elsewhere [14].The scenario of optimal unambiguous discrimination of two density operators is as follows: two (normalized) density operators ̺ 1 and ̺ 2 , acting on a finite-dimensional Hilbert space H occur with a priori probability p 1 and p 2 , respectively, where p 1 + p 2 = 1. We will denote * Electronic address: Matthias.Kleinmann@uibk.ac.at the support of a density operator ̺ as the orthocomplement of its kernel, (supp ̺) ⊥ = ker ̺. A measurement for USD is described by a positive operator valued measure (POVM), i.e., a family of positive semi-definite operators {E 1 , E 2 , E ? } with E 1 + E 2 + E ? = 1 1, obeying the constraints for unambiguity, tr(E 2 ̺ 1 ) = 0 and tr(E 1 ̺ 2 ) = 0. The operator E ? corresponds to the inconclusive outcome while E 1 and E 2 correspond to the successful detection of ̺ 1 ...