This paper describes an approach based on mathematical morphological reconstruction for extraction of a tumor from labeled brain magnetic resonance imaging (MRI) images affected by impulse noise. As a part of pre-processing, global thresholding technique is applied on the input MRI image. Afterward, mathematical morphological reconstruction operation segments a brain tumor from the processed image. Algorithm is further modified for non-uniform intensity regions of the brain tumor. Optionally, mathematical morphological operator is utilized to remove pepper noise if presents in tumor region. Finally, tumor region is extracted using resultant image and input MRI image. Experimental results show the proposed method effectively extracts a brain tumor from noisy labeled MRI images.