Provided is a tire defect detection method capable of accurately detecting a thinly extending convex defect of a tire surface. Prior to the start of Step S1, two-dimensional images including a slit light image are successively obtained in advance. In Step S1, a slit light image is extracted from data of a plurality of shot two-dimensional images. In Step S2, an eccentricity component which is deviation resulting from eccentricity is eliminated from the extracted slit light image. In Step S3, a feature quantity is calculated based on the light image from which the eccentricity component is eliminated, and in Step S4, a thinly extending convex defect is detected based on the calculated feature quantity.