We propose a fundamental frequency (F0) estimation method which is fast, accurate and suitable for real-time use. While the proposed method is based on the same framework as DIO [1, 2], it has two clear differences: it uses RMA (Recursive Moving Average) filters for attenuating high order harmonics, and the period detector is designed to work well even for signals which contain some higher harmonics. Effect of trace-back duration of post-processing was also examined. Evaluation experiments using natural speech databases showed that the accuracy of the proposed method was better than DIO, SWIPE' [3] and YIN [4] and computation speed was the fastest compared to those existing methods.