Children with autism spectrum disorders (ASD) often emit errors during the establishment of conditional discriminations. These children may not respond to more traditional error-correction procedures, such as least-to-most prompting. In this study, we compared two other types of error-correction procedures, namely embedding an identity-matching task as a differential observing response (DOR; Fisher et al. in J Appl Behav Anal 40:489-499, 2007) compared with a second-order fixed ratio 3/fixed ratio 1 (FR3 FR1) schedule with response cost procedure (Fisher et al. in J Appl Behav Anal 47:738-748, 2014). Results garnered from a multiple baseline design with embedded adapted alternating treatment design components demonstrated that the identity-matching/DOR procedure lead to superior acquisition of conditional discriminations for two boys with ASD. These findings suggest that for audio-conditional discrimination training, the identity-matching/DOR method may be more effective over the second-order FR3 FR1 with response cost approach.