Machine vision technology is widely used in industrial measurement. To realize high precision measurement of machine vision is of great significance to precision machining and manufacturing. The problem that measuring object will cause measurement error in machine vision application is studied when it deviates from the focal plane or has a certain thickness. The influence of parallelism error caused by lens telecentricity on measurement error of defocus sample is discussed on emphasis. The experimental results show that among all measurement errors caused by sample deviation from the optimal imaging plane, the proportion of parallelism error accounts for about 90% . Measuring accuracy can be improved greatly by compensating parallelism error. Additionally, in order to analyze the problem that the edge is fuzzy when sample has a large thickness, several edge detection algorithms are used. The results show that within a certain range, the thicker the object is, the larger the edge detection error is. On the basis of this result, a compensation method based on image gray-level curve is put forward. As a result, measurement error has decreased from more than 20 μm to less than 10 μm.