Hop timing estimation for noncooperative frequency-hopped (FH) signals is an important problem that feeds into the multiple facets of military communications, from interception of noncooperative communications to interference mitigation. If there is a mismatch between the receiver's observation band and the hop bandwidth of the transmitted signals in an FH system, signals appear to hop in and out of the observation frequency band. This causes model-order variations and high-resolution parametric methods that assume fixed model order may fail. Using an antenna array and capitalizing on the data structure of 2-D frequency mixture (space and time), this paper proposes an expectation-maximization (EM) algorithm for blind hop timing and frequency estimation with possible bandwidth mismatch. The initialization of the EM algorithm is given by the data spectrogram. Simulation results demonstrate that the proposed method performs well even at low-to-moderate signal-to-noise ratio with unknown hop sequences.