In asynchronous motors, colored Gaussian noises will be produced in stator current after lowpass filter. For this reason, the precision of rotor fault detection methods was restricted. In order to solve this problem, a method for the rotor fault detection in asynchronous motors which could restrain colored Gaussian noises was proposed. Firstly, inverse rotation transform was used to pretreat the sampling stator current and eliminate the fundamental frequency, with which the estimation error of direct detection without pretreatment could be avoided. Then, considering that crosscorrelation function (CCF) could restrain colored Gaussian noises, crosscorrelation function Hankel total least squares (CCFHTLS) algorithm was proposed to detect motor rotor fault. The experimental result of broken rotor bars and eccentric fault detection indicated that CCFHTLS algorithm remarkably restrained colored Gaussian noises and accurately preserved available information of faults, significantly promoting the detection precision.