Cross-domain fault diagnostic methods based on domain adaptation (DA) have been developed for single-sensor monitoring scenarios, in which the source and target domains fall into the same categories. However, in real-world situations, faults are usually mixed with each other, and the target health category is a subspace of the source health category, posing challenges to the current cross-domain fault diagnostic approaches. Additionally, with the increasing complexity of modern industrial systems, less attention has been paid to multisensor cross-domain diagnosis. To address this research gap, this article proposes a new method of multisensor partial DA fault diagnosis. First, the frequency information of multisensor measurements is obtained to fully utilize the fault information. Then, an improved partial DA method based on a weighted domain adversarial network is used to distinguish the label space of the data samples. Finally, a joint optimization objective is constructed under the framework of partial transfer fault diagnosis, where two terms, namely, conditional entropy and adaptive uncertainty suppression, are further added to regularize the optimization objective. Through the proposed method, the positive transfer between shared common classes is guaranteed, and additionally, the passive influence resulting from outlier classes is prevented. Experimental results show that the proposed approach surpasses other popular methods for partial transfer fault diagnosis.