Bearing defective inspection plays a vital role in bearing quality control. Unlike signals in the process of condition monitoring andfault diagnosis, the signal characteristic of defective bearings is much weaker and difficult to be quantified through the accelerationbased techniques. In this paper, a novel system is developed to inspect automatically the small defects of roller bearings for on-linequality control. Rather than using acceleration based techniques the system employs a high sensitive eddy current sensor to measurethe displacement profiles of the outer race for high signal to noise ratio. Furthermore, a morphological filter is used to enhance thefeature signal which is subsequently measured by Kolmogorov complexity measure. Both simulated signals and measured data showthat this system is able to diagnose defects including abnormal surface roundness, waviness, misaligned races which are typicalquality problems in bearing manufacturing lines.