Accurate and efficient identification of bat (Microchiroptera) echolocation calls has been hampered by poor knowledge of the intraspecific variability in calls (including regional variation), a lack of call parameters for use in separating species and the amount of time required to manually identify individual calls or call sequences. We constructed and tested automated bat call identification keys for three regions in New South Wales, Australia, using over 4,000 reference calls in ≈300 call sequences per region. We used the program AnaScheme to extract time, frequency and shape parameters from calls recorded with the Anabat system. Classification trees were built to separate species using these parameters and provided the decision rules for construction of the keys. An ‘Unknown’ category was included in the keys for sequences that could not be confidently identified to species. The reliability of the keys was tested automatically with AnaScheme, using independent sets of reference call sequences, and keys were refined before further testing on additional test sequences. Regional keys contained 18–19 species or included species groups. We report rates of sequence misidentification (accuracy) and correct identification (detection) relative to all sequences (including ‘unknowns’) used to test each version of a key. Refined versions of the keys were accurate, with total misidentification rates of 0.5–5.3% for the three regions. Additionally, total correct identifications for regions were 56–75% (> 50% for most species), an overall high rate of detection. When ‘unknowns’ were ignored, as is common in many published studies, correct identification for regions increased to 91–99%, rates which compare favourably to the most successful classifiers tested to date. The future use of AnaScheme for automated bat call identification is promising, especially for the large-scale temporal and spatial acoustic sampling to which Anabat is particularly suited.