Spatial heterogeneity is an important characteristic of the land surface. Because multi-spectral bands are used to describe the land surface, an approach has to be established to characterize the surface spatial heterogeneity from multi-spectral remote-sensing observations. This work aims at quantifying the spatial heterogeneity of cropland using varigorams for multi-temporal NDVI, near infrared (NIR) and red reflectance. A concept of mean length variability is proposed to compare the difference in spatial heterogeneity detected by variables with different magnitudes. The important temporal changes in spatial heterogeneity observed by NDVI, NIR and red bands over cropland are a result of changes in the fraction of vegetation cover. The results indicate the following: (1) the NIR and red variables detect a similar spatial heterogeneity of the cropland with similar values of the mean length variability before the sowing of crops; (2) the NDVI, NIR and red values capture different degrees of spatial heterogeneity when vegetation cover is low; (3) over medium vegetation cover, the NDVI and NIR values capture similar spatial heterogeneity, which is low compared to the red band due to the homogeneity of soil; and (4) the spatial heterogeneity quantified by the NIR value is more heterogeneous than those of the NDVI and red values when vegetation cover is high. The red reflectance is sensitive to soil properties while the NIR reflectance responds to vegetation. The spatial heterogeneity of red reflectance decreases and that of the NIR reflectance increases with the growth of vegetation. The NDVI value shows the greatest heterogeneity in the early stage of crop growth. With an increase in the image pixel size, the spatial heterogeneity quantified by the mean length variability of the NDVI, NIR and red variables tends to be the same.