Abstract: When constructing an unsupervised adaptive classifier based on extreme learning machines (ELMs), the hidden layer parameters are usually randomly selected, which do not have domain adaption capability. In order to enhance the knowledge transfer ability of the cross-domain ELM, a new unsupervised domain adaptive classifier learning method via ELMs is proposed. The method mainly uses the extreme learning machine autoencoder to reconstruct the data of both source and target domains, which can obtain the hidden layer parameters with domain invariant features. Furthermore,by using the ideas of joint distribution adaption and manifold regularization, the output layer weights of ELMs can beadaptively adjusted. The proposed classifier can provide the domain adaption capabilityto both hidden layer and outputlayer parameters of ELMs. Experiments on the digital and object recognition datasets show that the proposed classifierhas higher cross-domain classification accuracy.