This paper proposes an adaptive edge detection technique based on multi-scale analysis of odd Gabor transform. In order to obtain the odd Gabor transform based edge response efficiently, an improved scheme is suggested, which can avoid the inverse influence brought by the estimation error of gradient direction on the construction of odd Gabor filter templates. To sharpen the edge response, an adjacent scales multiplication in odd Gabor transform domain is introduced. Moreover an adaptive threshold determining method related to the exponential distribution property of sharpened edge response is also presented. The experimental results on both synthetic and real world images show that our scheme is robust and has good edge detection performance.