The fingerprints difference between the individual transmitters is so subtle and can be seriously affected by noise, and the labeled training data samples are difficult to obtain especially in non-collaborative conditions. To solve these problems, we propose a transmitter fingerprint feature recognition method based on semisupervised collaborative representation classifier (CRC). The envelope properties of the boot signal are adopted as the fingerprint features of individual transmitters. In order to reduce the effect of ambient noise, a noise suppression method is given based on higher order cumulant. And finally the semi-supervised CRC is constructed to classify the feature results. Experiments demonstrate that our method could suppress the noise effectively, and have higher recognition rate by making effective use of the data features of unlabeled samples.