This paper presents a computationally efficient FIR (finite impulse response) filter for ECG (electrocardiogram) baseline wander removal. The proposed filter is based on frequency-response masking (FRM) technique, which reduces the number of multiplications by 90% compared to the conventional FIR filter. The filter specifications are optimized to obtain the highest SNR (signal-to-noise ratio) while minimizing the number of filter coefficients. The SNR performance of the proposed filter is evaluated using ECG signals from MIT-BIH Arrhythmia Database with added baseline noises from MIT-BIH Noise Stress Test Database. Simulations show that the FRM filter is capable of removing most of baseline wander while preserves ECG signal.