Accelerator mass spectrometry (AMS) of 36Cl ( t 1/2 = 0.30 Ma) at natural isotopic concentrations requires high particle energies for the separation from the stable isobar 36S and was so far the exclusive domain of tandem accelerators with at least 5 MV terminal voltage. Using terminal foil stripping and a detection setup consisting of a split-anode ionization chamber and an additional energy signal from a silicon strip detector, a 36S suppression of >10 4 at 3 MV terminal voltage was achieved. To further increase the 36S suppression energy loss straggling in various counter gases (C 4H 10, Ar–CH 4 and C 4H 10–Ar) and the effect of “energy focusing” below the maximum of the Bragg curve was investigated. The comparison of experimental data with simulations and published data yielded interesting insights into the physics underlying the detectors. Energy loss, energy straggling and angular scattering determine the 36S suppression. In addition, we improved ion source conditions, target backing materials and the cathode design with respect to sulfur output and cross contamination. These changes allow higher currents during measurement ( 35Cl − current ≈ 5 μA) and also increased the reproducibility. An injector to detector efficiency for 36Cl ions of 8% (16% stripping yield for the 7+ charge state in the accelerator, 50% 36Cl detection efficiency) was achieved, which can favorably be compared to other facilities. The memory effect in our ion source was also thoroughly investigated. Currently our measured blank value is 36Cl/Cl ≈ 3 × 10 −15 when samples with a ratio of 10 −11 are used in the same sample wheel and 36Cl/Cl ≈ 5 × 10 −16 if measured together with samples with a ratio of 10 −12 or below. This is in good agreement with the lowest so far published isotope ratios around 5 × 10 −16 and demonstrates that 3 MV tandems can achieve the same sensitivity for 36Cl as larger machines.