Water capture mechanisms of zeolitic imidazolate framework ZIF-90 are revealed by differentiating the water clustering and the center pore filling step, using vibrational sum-frequency generation spectroscopy (VSFG) at a one-micron spatial resolution and state-of-the-art molecular dynamics (MD) simulations. Through spectral lineshape comparison between VSFG and IR spectra, the relative humidity dependence of VSFG intensity, and MD simulations, based on MB-pol, we found water clustering and center pore filling happen nearly simultaneously within each pore, with water filling the other pores sequentially. The integration of nonlinear optics with MD simulations provides critical mechanistic insights into the pore filling mechanism and suggests that the relative strength of the hydrogen bonds governs the water uptake mechanisms. This molecular-level detailed mechanism can inform the rational optimization of metal-organic frameworks for water harvesting.