We report the use of frequency compounding to significantly reduce speckle noise in optical coherence microscopy, more specifically on the en face images. This method relies on the fact that the speckle patterns recorded from different wavelengths simultaneously are independent; hence their summation yields significant reduction in noise, with only a single acquisition. The results of our experiments with microbeads show that the narrow confocal parameter, due to a high numerical aperture objective, restricts the axial resolution loss that would otherwise theoretically broaden linearly with the number of optical frequency bands used. This speckle reduction scheme preserves the lateral resolution since it is performed on individual A-scans. Finally, we apply this technique to images of fixed human brain tissue, showing significant improvements in contrast-to-noise ratio with only moderate loss of axial resolution, in an effort to improve automatic three-dimensional detection of cells and fibers in the cortex.