The objective of this study was to establish reliability estimates of different versions of the Finger-to-Nose test (FNT) using the generalisability theory. Generalizability theory was used to estimate the relative contribution of trials and occasions to the total measurement error and to generate reliability coefficients. Forty-two healthy adults (aged 20-63 yrs old) free of arm pain or other pathology participated. Participants performed 10 versions of the FNT during two evaluation sessions separated by a 24-hour interval. The time (s) to perform five trials of each version was recorded. The mean reliability coefficient for the one-occasion/one-trial testing situation among the test versions was 0.79 (range = 0.77-0.82). For all versions, the principal source of variance of the test performances came from the participants; minimally important sources of measurement error were also associated with occasion and the interaction effects between participants and occasion. Number of trials was not found to be an important source of error. The 10 versions of the FNT under investigation provide reliable measures of upper limb coordination among healthy adults when assessed by the one evaluator generalised across two days. These results are important to the development of norms for this measure.