Caffeine is the most widely used psychostimulant in the world, though preclinical studies suggest weaker evidence for abuse-related effects than stimulants with high abuse liability, such as amphetamine or cocaine. Intracranial self-stimulation (ICSS) is one procedure used to assess the abuse liability of drugs, and previous studies have produced mixed results regarding whether caffeine produces an abuse-related facilitation of ICSS. This study assessed both caffeine and its main metabolite in humans, paraxanthine, using a frequency-rate ICSS procedure and compared their effects to those of amphetamine and cocaine. Male Sprague-Dawley rats were implanted with intracranial electrodes targeting the medial forebrain bundle and trained to respond under a fixed-ratio 1 schedule for brain stimulation that varied across a range of frequencies (56–158 Hz in 0.05 log increments). Data analysis focused on three dependent measures: reinforced responding (defined as responses that produced brain stimulation), non-reinforced responding (defined as responses that occurred during each 0.5 sec brain stimulation and that did not produce additional stimulation), and total responding (reinforced + non-reinforced responding). Both amphetamine and cocaine produced robust increases in total, reinforced and non-reinforced responses. Caffeine also increased total, reinforced and non-reinforced responses, but the caffeine dose-effect curve had an inverted-U shape, and peak ICSS facilitation was less than that produced by amphetamine or cocaine. Paraxanthine increased only total responses and non-reinforced responses. These results suggest that paraxanthine has low abuse liability and does not mediate abuse-related effects of caffeine.