Maximum a posteriori (MAP) sequence detection for channels with intersymbol interference (ISI) has previously required knowledge of the channel sampled impulse response (SIR). Generally the SIR coefficients are determined via least mean square (LMS) or recursive least squares (RLS) estimation algorithms. For many unguided media channels such as mobile radio and high frequency radio which exhibit a time-varying SIR, these estimators must be adaptive. Adaptive estimators often fail to track adequately and are a major source of detector deterioration. A novel, blind maximum likelihood sequence detection (BMLSD) formulation without the need for external channel SIR estimation is proposed. The BMLSD performance is evaluated via simulation over several fast Rayleigh fading channels, which indicates substantial improvement compared to the conventional MLSD.