<p>The accurate interpretation of magnetic anomaly data relies heavily on knowledge of the total magnetization direction of subsurface sources. Particularly when strong remanent magnetization is present, assessing the total magnetization direction is crucial for interpreting magnetic anomalies. However, magnetic field observations are prone to various noises. This study aims to enhance the accuracy of magnetization direction determination by leveraging transformed fields that are less sensitive to noises. Our primary goal is to develop a robust method to estimate the magnetization direction, especially when dealing with complex magnetic anomalies. We propose a method for determining magnetization direction by exploiting the correlation between the pseudo-gravity (PG) field (which we demonstrate is less affected by noises) and the Normalized Source Strength (NSS). This involves searching for the NSS field derived from the observed magnetic field data and the PG fields computed from a range of assumed magnetization directions. The proposed method is validated through both synthetic modeling experiments and real data applications. The results demonstrate the robustness of the method in handling diverse anomaly geometries, including cases with mutual interference and random noises. Furthermore, it effectively mitigates the influence of remanent magnetization. The experiment on sphere source with 5%, 10%, and 20% random noises showed that our method yields significantly lower errors in estimating magnetization direction than previous approach. The mean errors for inclination and declination are 0.29°, 1.22°, 2.45° and 0.44°, 0.84°, 1.92°, respectively. Consequently, our approach, utilizing the NSS field and the PG field, offers an effective tool for estimating magnetization directions.</p>