Abstract Loss in axial compressor bleed systems is quantified, and the loss mechanisms identified, in order to determine how efficiency can be improved. For a given bleed air pressure requirement, reducing bleed system loss allows air to be bled from further upstream in the compressor, with benefits for the thermodynamic cycle. A definition of isentropic efficiency which includes bleed flow is used to account for this. Two cases with similar bleed systems are studied: a low-speed, single-stage research compressor and a large industrial gas turbine high-pressure compressor. A new method for characterising bleed system loss is introduced, using research compressor test results as a demonstration case. A loss coefficient is defined for a control volume including only flow passing through the bleed system. The coefficient takes a measured value of 95% bleed system inlet dynamic head, and is shown to be a weak function of compressor operating point and bleed rate, varying by ±2.2% over all tested conditions. This loss coefficient is the correct non-dimensional metric for quantifying and comparing bleed system performance. Computations of the research compressor and industrial gas turbine compressor identify the loss mechanisms in the bleed system flow. In both cases, approximately two-thirds of total loss is due to shearing of a high-velocity jet at the rear face of the bleed slot, one quarter is due to mixing in the plenum chamber and the remainder occurs in the off-take duct. Therefore, the main objective of a designer should be to diffuse the flow within the bleed slot. A redesigned bleed slot geometry is presented that achieves this objective and reduces the loss coefficient by 31%.