This paper investigates the accuracy of recall data by comparing administrative records with retrospective, self-reported survey responses to income and asset questions for a sample of self-employed households from coastal India. It finds that the magnitude of the recall error increases over time, in part because respondents rely less on memory and instead infer earnings based on past earnings. Individuals tend to recall monthly earnings more accurately when they are higher than the median. These results imply that the variance estimated from the self-reported earnings distribution will be lower than the real one. The paper also finds that data reported by income earners are more accurate than those by their wives. In addition, the use of time cues can worsen accuracy if they are not relevant to the respondent. Where the recall questions are placed in the two-hour long survey, however, does not affect accuracy.