By using an on-site collapse event as a reference and eliminating path effects of seismic-wave propagation, we constrain source characteristics of North Korea’s 3 September 2017 nuclear test. In the framework of a complete source model for an underground explosion consisting of an isotropic source representing the explosion and two secondary sources accompanying the explosion, with one represented by a double couple (DC) attributed to tectonic release and the other an oblique compensated linear vector dipole (CLVD) attributed to source medium damage, the source model of the 2017 test is inferred from its noncircular radiation patterns of Rayleigh and Love waves using seismic observations of the collapse as references. Based on the opposite azimuths of the inferred oblique CLVD of the test and the collapsing force of the event 8.5 min later, we propose a deformation scenario of the 2017 test in that the test generates a prompt source medium damage expanding away from the test center mostly along the direction of 320°(represented by the oblique CLVD) and the deformed surrounding rocks collapse toward the test center along the weakness direction in an opposite way 8.5 min later (represented by the collapse). We further propose that an inferred obliqueness of CLVD accompanying a nuclear test may serve an indication of possible collapsing direction for the test. Because the surface‐wave magnitude is strongly contaminated by secondary sources, we estimate the yield of the 2017 test to be 109.4±48.6kt based on the body‐wave Lg magnitude.