Perceptual learning studies in vision attribute patterns of specificity to learning in retinotopically-organized areas, with difficult tasks being more likely to show specificity when trained sequentially with a single point of transfer (Ahissar & Hochstein, 1997). Liu and Vaina (1998) developed an assay for testing specificity in which training on a base condition and the transfer condition are interleaved in uneven proportions. For motion direction discrimination tasks with different base directions, they showed substantial task transfer. Using a similar test design, we trained subjects in a 2AFC orientation discrimination task using Gabor patches in noise, where the discrimination was relatively difficult (with conditions +/-5∞ from a reference angle). Trials were interleaved in uneven proportions using the following schedule, A-A-B-A-A-B... 'A' trials were presented on one diagonal while 'B' trials were presented on the opposite diagonal with opposite reference angles. If learning were independent for the two types of stimuli, performance on the first half of 'A' trials would equal the performance on 'B' trials. If learning transfers between 'A' and 'B' trials, the performance on 'B' trials would be higher than the first half of 'A' trials, but lower than the second half of 'A' trials. Although interleaved designs might encourage transfer, we found that the learning in 'B' trials exactly equated to the learning in the first half of 'A' trials, indicating little transfer, or full specificity of learning in this difficult task. This extends claims about lack of transfer in difficult tasks to interleaved training protocols.