Recent experiments showed that the alignment-dependent ionization probabilities of hydrogen molecules vary as a function of laser intensity, and the anisotropy deviates from the prediction of the molecular tunneling ionization model (MO-ADK). To investigate the physical origins of the deviation, we systematically studied the anisotropy of hydrogen molecules in intense laser fields for three wavelengths, 400 nm (multiphoton ionization), 1850 nm (tunneling ionization), and 800 nm (in between), by solving the time-dependent Schrodinger equation with a model potential. The calculated ratio of the ionization probabilities for laser fields parallel and perpendicular to the molecular axis are in reasonable agreement with experiment. Furthermore, by analyzing the molecular wave function and the model potential, we found that the discrepancies between experiment and the MO-ADK prediction originated from the inaccurate coefficients used in the model and that the assumption of isotropy of the effective potential in the tunneling region is invalid.