Abstract: Cultivated land, as an important agricultural resource, being quickly and accurately extracted using remote sensing is the key step for its protection and sustainable use. An approach was presented based on multiscale fractal signature and image threshold segment to cultivated land extraction from ETM+ remote sensing images. The blanket method used for computing fractal dimension of surface was modified to calculate fractal dimension of curves including the “up” and “down” fractal signatures at each scale based on the spectral curve of each pixel. The computed fractal signatures of the cultivated land and other land covers at corresponding scale were analyzed and compared; The results showed that cultivated land was enhanced through differencing between the third and fourth down fractal signature, and its spatial distribution was further extracted using image segmentation. The segmentation area accuracy and point positioning accuracy were as high as 91.0% and 71.8%, respectively. The point positioning accuracy of the rainfed cultivated land was relatively low, while irrigated cultivated land and vegetable land were up to 86.6%. The method developed in the study can effectively extract cultivated land and provide a reference for further related researches.