This study examined changes in recall accuracy for mobile phone calls over a long period. Japanese students' actual call statuses were monitored for 1 month using software-modified phones (SMPs). Three face-to-face interviews were conducted to obtain information regarding self-reported call status during the monitoring period: first interview: immediately after the monitoring period; second interview: after 10-12 months; third interview: after 48-55 months. Using the SMP records as the "gold standard", phone call recall accuracy was assessed for each interview. Data for 94 participants were analyzed. The number of calls made was underestimated considerably and the duration of calls was overestimated slightly in all interviews. Agreement between self-report and SMP records regarding the number of calls, duration of calls and laterality (i.e., use of the dominant ear while making calls) gradually deteriorated with the increase in the interval following the monitoring period (number of calls: first interview: Pearson's r=0.641, third interview: 0.396; duration of calls: first interview: Pearson's r=0.763, third interview: 0.356; laterality: first interview: weighted-κ=0.677, third interview: 0.448). Thus, recall accuracy for mobile phone calls would be consistently imperfect over a long period, and the results of related epidemiological studies should be interpreted carefully.