This paper studies blind constrained minimum output energy (CMOE)-based and subspace-based linear minimum mean-squared-error (LMMSE) detectors for multi-carrier code division multiple access (MC-CDMA) systems. By imposing quadratic weight constraint, the CMOE detector is made more robust against signature waveform mismatch, and a better performance over the standard CMOE detector is obtained. Because of separation of signal and noise subspaces, the more complicated subspace-based LMMSE detector has better performance than the CMOE detector. The recursive subspace tracking algorithms are also investigated for the subspace-based MMSE receiver. Numerical results show that the steady-state performance of the robust CMOE detector is close to the subspace-based MMSE method. The blind mode decision-directed LMMSE detection is studied where the blind detectors are used for initial adaptation. Numerical simulations illustrate that the blind mode decision-directed MMSE detection substantially improves the system performance when the frequency-selective channel is slowly-varying