Abstract: ‍ ‍In cell-free massive multiple-input multiple-output networks， the characteristics of channel hardening and favorable propagation caused by distributed deployment of access points （APs） are no longer significant. At the same time， malicious eavesdroppers who perform pilot spoofing attack （PSA） can simulate the channel characteristics by adjusting the pilot transmission power. Therefore， it is invalid for PSA detection schemes based on large-scale fading prior information. The PSA detection based on the information theory criterion requires more pilot samples than the number of APs， and the time occupied by the channel estimation has an impact on spectral efficiency. In view of these situations， a signal subspace based spoofing detection （SSBSD） method is proposed. First， random sequences are superimposed on the training sequences of legitimate users， and the number of sources is estimated by flexible detection criterion. Secondly， by exploiting Linear Shrinkage in Statistics， the covariance matrix of pilot samples is optimized to make the eigenvalues ​​of samples have a better approximation of the overall distribution. Finally， the samples are ​​projected into the orthogonal complementary space of the signal to identify the wiretapped user. After simulation experiments and a theoretical analysis of the SSBSD method， it is not necessary to a priori of large-scale fading， and the same pilot length as for APs. It is very sensitive to the transmission power of PSA pilots. In communication environments where large-scale fading is unknown， large-scale fading is time-dependent， or spectrum resources are limited， the PSA at a low level of power can be accurately detected.