The ultraviolet (UV) sensitivity is determined by a phototest where the skin is exposed to well-defined doses of UV radiation and the resulting erythema is graded by visual scoring after 20-24 h. In this study we wanted to estimate the reproducibility of erythema assessment in phototesting. Twenty-one healthy Caucasians with skin types I to IV were phototested on UV un-exposed buttock skin using a xenon lamp solar simulator. Twenty-four hours after UV exposure eight physicians independently graded the erythema reactions two times. Data were analysed using inter- and intra-observer agreement and kappa statistics, which adjusts for agreement that could be caused by chance alone. Observed agreement and kappa statistics were found to decrease with increasing intensity of erythema and to be lower for skin types III and IV compared to skin types I and II. Intra-observer agreement was uniformly better than inter-observer agreement. The difference between observers assessment could be as much as three clinical erythema grades. Physicians's previous experience with phototesting only had a minor influence on agreement. In conclusion, phototesting is based on subjective assessment of erythema and is not as precise and reproducible as expected. Agreement was better for barely perceptible erythema than for erythema with a well-defined border and we therefore recommend that the barely perceptible erythema reaction should be used for measurement of the minimal erythema dose.