This paper investigates whether modulation domain speech en-hancement methods are better than corresponding acoustic do-main methods when used as a preprocessor to automatic speech recognition. It is well known that linguistic information of speech is contained not only in the short-time magnitude spec-trum but also in its temporal evolution. In addition, this study investigates whether popular metrics used in speech enhance-ment (such as PESQ, segmental SNR, STOI) are indicative of ASR performance. ASR experiments on the TIMIT speech cor-pus corrupted by various noises were performed to compare re-cent modulation domain methods with their acoustic domain variants.