This paper presents the findings of an effort undertaken by the Federal Highway Ad ministration (FHWA) Long-Term Pavement Performance (LTPP) program in the United States of America (USA) to assess variability of distress data. Two procedures were used on collecting the pavement distress data, the manual procedures in the field and photographic method.The primary source of data used to characterize the variability of manual procedures are the results from nine LTPP distress rater accreditation workshops conducted between 1992 and 1996. This data source provided the distress ratings of 6 to 16 individual raters per workshop who performed the ratings on the same day on the same test section; each workshop included an asphalt concrete (AC) and a Portland cement concrete (PCC) pavement test section. In ad dition, a reference rating of the distress present on the same test sections was performed using a consensus rating method by a group of expert raters immediately prior to the workshop; data from these reference surveys were used as a surrogate of "ground-truth". Based on the analysis of these data, both the apparent bias and precision for the common distress type-severity level combinations were quantified.Data used for evaluating the film-derived distress data variability are the results of distress survey conducted on 12 pavement test sections, including six asphalt concrete (AC) and six Portland cement concrete (PCC) pavements. Distress surveys were performed by experts, indi vidual raters, and two-person teams. A consensus survey on the pavements by a group of ex perts based on the photographic method was also conducted. Finally, as in the manual survey case, a reference rating of the distress present on the same 12 test sections was performed using a consensus manual rating method by the same group of experts; data from these reference sur veys were used as a surrogate of "ground-truth".Based on the analysis of these data, variability of the film derived distress data was evalu ated and both apparent bias and precision for the data were quantified. It was observed that da ta interpreted by two-person teams were more consistent and showed smaller variations.