<p>Fatigue failure in roller bearings constitutes a critical bottleneck for the safety and reliability of modern high-speed, heavy-load, and intelligent machinery. To address this challenge, this review systematically constructs a comprehensive, multi-scale analytical framework for bearing fatigue failure, spanning from microscopic to macroscopic influences. We comprehensively analyze the complex, coupled interplay of three core factors—structural parameters, material properties, and operational conditions—in the initiation and propagation of fatigue cracks. At the structural level, this review focuses on the critical influence of roller logarithmic profiles, surface microtextures, roller dimension deviations, and preload/clearance settings on contact stress distribution. On a material level, we systematically examine how bearing steel microstructures, non-metallic inclusions, and advanced surface hardening processes (e.g., cryogenic treatment, ultrasonic rolling strengthening) fundamentally regulate anti-fatigue performance. Furthermore, we emphasize how dynamic operating conditions such as temperature increases, preload deviations, lubrication failures, and contaminants collectively accelerate fatigue damage. Recognizing the limitations of traditional methods in capturing degradation under these nonlinear and dynamic conditions, this review highlights recent advances in remaining useful life prediction and structural optimization using multi-source signal fusion, deep learning, and intelligent algorithms. This hybrid "physics-constrained and data-driven" paradigm offers a new approach for achieving robust and highly accurate life assessments. The aim of this review is to provide fundamental theoretical insights into the failure mechanisms of bearings under complex conditions and to offer practical technical support for the design and maintenance of next-generation, high-reliability machinery.</p>