Using data collected with the HyperCP (E871) spectrometer during the 1997 fixed-target run at Fermilab, we report the first observation of the decay K − → π − µ + µ − and new measurements of the branching ratios for K ± → π ± µ + µ − . By combining the branching ratios for the decaysThe CP asymmetry between the rates of the two decay modes is [Γ(PACS numbers: 13.20. Eb, 14.40.Aq, 11.30.Er The rare decay of charged K mesons to a pion and a lepton pair (K ± πll , where l = e or µ) can be used to study flavor-changing neutral currents as a higher-order process in the standard model and to explore new physics. However, to realize these goals, it is necessary to understand the dominant K ± → π ± γ * radiative transition which involves long-distance hadronic effects. Since it is difficult to calculate such effects, chiral perturbation theory (ChPT) including electroweak interactions has been applied to K + πll [1]. In a recent model-independent analysis with ChPT at O(p 6 ) [2], the K + πll decay rate and the form factor characterizing the dilepton invariant-mass spectrum are calculated in terms of two parameters a + and b + . Determining these parameters from the measured branching ratio and the dilepton mass spectrum of the K + πee decay, this analysis predicts the ratio R = B(K + πµµ )/B(K + πee ) to be greater than 0.23 and an increase in the CP asymmetry between the decay rates of K + πll and K − πll compared to the leading-order estimation in the chiral expansion [3]. The resulting expected CP asymmetry is ∼ 10 −5 [2, 3].The K + πee decay has been studied by several experiments [4,5,6], and the Particle Data Group (PDG) has compiled a mean branching ratio B(K + πee ) = (2.88 ± 0.13) × 10 −7 [7]. The K + πµµ decay was first observed by the E787 Collaboration at the Brookhaven National Laboratory (BNL). With 13 fully reconstructed 3-track and 196 partially reconstructed 2-track events, they determined B(K + πµµ ) = [5.0 ± 0.4(stat) ± 0.7(syst) ± 0.6(theor)] ×10 −8 [8]. The E865 Collaboration at BNL has subsequently observed 430 fully reconstructed K + πµµ events and measured B(K + πµµ ) = [9.22 ± 0.60(stat) ± 0.49(syst)] ×10 −8 [9]. The discrepancy between these two experimental results is more than three standard deviations and is not understood. In addition the value of R obtained with the E787 result and the average value of B(K + πee ) is inconsistent with the prediction and is difficult to accommodate within the standard model [2]. It is thus important to resolve the discrepancy in the B(K + πµµ ) measurements.