The invention relates to a weighted iterated low-field nuclear magnetic resonance T2 spectrum inversion algorithm which comprises the following steps of (1), reading an original data file which is obtained through acquisition by low-field nuclear magnetic resonance equipment; (2), performing preprocessing operation on the original data for obtaining an inversion core matrix K and a vector m whichis composed of signal amplitudes of echo peaks in the original data at an echo time; (3), calculating signal-to-noise ratio, curvature and gradient of the acquired data, and furthermore acquiring an inversion weight matrix; (4), solving a formula m=Ks through weighted iteration of the inversion weight matrix, obtaining an optimal solution of s, wherein s represents a vector which is composed of acontent of a material which corresponds with transverse relaxation time; and (5), drawing a T2 spectrum according to the optimal solution. Compared with prior art, the algorithm is advantageous in that short relaxation components are weighted, thereby better approaching to sparsity of an original signal, and obtaining high calculation precision and high robustness; and on the condition of different signal-to-noise ratio data, a stable inversion result can be obtained.