A large-scale antenna array introduces the innegligible propagation delay for a received signal across the array aperture in addition to a phase rotation. If the delay is comparable to a symbol period in wideband millimeter-wave (mmWave) communications, then its impact on channel estimation and signal detection needs to be properly treated. In this case, different frequencies actually “see” distinct angles of arrival (AoAs) for the same physical path, which is also known as the beam squint effect. In this article, we propose a new channel estimation scheme with full consideration of beam squint for the mmWave massive multiple-input multiple-output (MIMO) systems. A super-resolution compressed sensing approach is first developed to jointly extract the initial AoA, the time delay, and the complex gain of each physical path, from which channel covariance matrix can be constructed rather than acquired through the long-term average. Then, the least-square (LS) and the linear minimum mean- squared error (LMMSE) channel estimators are designed with a small amount of training. Numerical results demonstrate the superiority of the proposed scheme over the conventional methods.