Abstract: In the application of multi-electromagnetic nondestructive testing technology in the testing of mechanical properties of ferromagnetic materials, due to ferromagnetic materials often show the characteristics of magnetic anisotropy under the influence of various factors, the test sample needs to keep a fixed direction during testing, the selection of detection direction becomes a problem worth studying. Based on tangential magnetic field harmonic analysis, Barkhausen noise detection, incremental permeability detection, multi-frequency eddy current detection and other methods, a circumferential multi-method electromagnetic non-destructive testing measurement system was built to explore the influence of different detection directions on mechanical property testing. A BP neural network prediction model was established after collecting electromagnetic parameters of cold-rolled ultra-high strength steel as an experimental object. K-fold cross-validation was used to evaluate the influence of detection direction on prediction accuracy. It was found experimentally that the distribution of circumferential electromagnetic characteristics of ultrahigh strength steel was not uniform, showing the characteristics of magnetic anisotropy, and the detection accuracy along the width direction of the experimental sample was better than that in the rolling direction.