In response to challenges such as weak early fault characteristic signals in rolling bearings under complex noise environments and the disconnection between condition monitoring and fault diagnosis, this paper proposes an integrated method for condition monitoring and fault diagnosis of rolling bearings based on a hybrid noise model and the maximum likelihood ratio (MLR). First, a statistical model of hybrid noise is constructed, and the expectation-maximization (EM) algorithm is used to estimate parameters and fit the vibration signals collected under healthy conditions, establishing this model as the health baseline; Next, the MLR index is constructed to quantify the probability distribution differences between the monitored signals and the health reference signals. On this basis, the exponential weighted moving average (EWMA) control chart is employed to process the sequence of health monitoring indicators, thereby amplifying the bearing′s degradation trend whilst retaining fault features. Finally, by reusing the band MLR evaluation index, the sub-bands after wavelet packet decomposition are screened to extract the optimal fault-sensitive frequency band. Subsequently, envelope spectrum analysis is performed on the signals of this frequency band to achieve fault diagnosis. Through experiments on two public datasets and comparative analysis with other methods, on the IMS dataset, the MLR index detects early faults 2.048 seconds earlier than the attention Lempel-Ziv complexity method and identifies an outer race fault characteristic frequency of 230 Hz. This result shows only about a 2.1% relative error compared to the theoretical fault characteristic frequency (235 Hz). On the XJTU-SY bearing dataset, the MLR index detects faults 1.28 seconds earlier than the effective weighted sparse kurtosis method and identifies two fault characteristics at 108 Hz and 175 Hz. These correspond to the theoretical outer race fault characteristic frequency (107 Hz) and inner race fault characteristic frequency (172 Hz), with relative errors of 1.4% and 1.7%, respectively. These results verify the timeliness and accuracy of the proposed method for integrated condition monitoring and fault diagnosis of rolling bearings.