The infinitely variable size of atmospheric particulate matter raises a huge challenge for accurately assessing the health risks of pollutants of different particle sizes. Currently, the majority of studies have concentrated on PM2.5-bound heavy metals (PM2.5-HMs) and their associated health risks, greatly ignoring the risks caused by heavy metals in other particle sizes such as PM10 and total suspended particulate (TSP). In this study, Pearson correlation analysis was used for the first time to explore the concentration correlation of 10 heavy metals (HMs) (Cd, Cr, As, Pb, Zn, Cu, V, Mn, Ni, Co) in PM1, PM2.5, PM10, and TSP, utilizing 91876 PM samples from China from 2000 to 2022. The results of the Pearson correlation analysis revealed that the correlation coefficients between PM2.5-HM and PM10-HM, PM10-HM and TSP-HM, and PM2.5-HM and TSP-HM were all above 0.82, with statistical significance (P < 0.001). Due to the sample size limitation of PM1-HM (n < 15), the Pearson correlation coefficients of PM1-HM with other particle-size HM may not be robust enough. Then generalized linear models (GLM) and generalized additive models (GAM) were used to exclude the interference of covariates (sampling time and sampling city), and 16 univariate inter-prediction models which are not affected by temporal and spatial factors for PM2.5, PM10, and TSP binding HM concentration were further constructed. The regression diagnostics of these prediction models indicate normally distributed data and strong linear trends, and the influential points in the modeled data were all within acceptable limits (n ≤ 1). The R2s of the prediction models were almost greater than 0.8, indicating that the prediction models were effective and robust. This study reveals for the first time the particle-size correlation of PM2.5-HM, PM10-HM, and TSP-HM, offering a novel approach for the prediction of heavy metal concentration and health risk assessment of atmospheric particle-size fractionated heavy metal.