The communication systems are susceptible to linear frequency modulation (LFM) interference especially when the radar and communication systems are on the same platform. This letter proposes an LFM interference suppression method for direct sequence spread spectrum (DSSS) communications via subspace projection without the requirement of interference parameter estimation. The subspace projection algorithm is successfully adapted to DSSS communication systems by constructing a data matrix that is composed of several certain delayed signals. We discussed the relationship between LFM interference and the data matrix, and derived the expression for the gain of this method. The time-varying eigenvalues are obtained by a robust recursive least squares learning algorithm (RRLSA) instead of matrix decomposition to reduce computational complexity. The simulation experiments show that this proposed method has excellent suppression effects and robustness on LFM interference.