It is widely held that certain parts of the auditory spectrum are more important than others for speech recognition. However, the concept of frequency importance for speech recognition has received relatively little attention from researchers over the years. This was probably because most assumed that published frequency importance functions for nonsense syllables were satisfactory for most applications. Recent research suggests that different types of speech materials have different importance functions and that predictions of speech recognition for one speech material cannot be made accurately under all conditions using an importance function for a different material. Research on this topic will be reviewed and some recent data presented. An important remaining question is the effect that size of the frequency importance function has under different listening conditions. Early findings suggest that, although the importance function has relatively little influence on outcomes under some conditions, it can be substantial in others.