High energy electron radiography (HEER) proposed first for real-time high spatial and temporal resolution diagnosis of warm dense matter (WDM) and inertial confinement fusion (ICF) has proved experimentally feasible for mesoscale sciences diagnosis. Until now, the spatial resolution of the images close to 1 μm has been reached experimentally which is better than that of X-rays and neutron radiography. However, traditional HEER obtains two-dimensional images which cannot accurately present the three-dimensional structure of the sample. To further improve the capability of HEER to diagnose and obtain the internal information of samples, three-dimensional high energy electron radiography (TDHEER) was put forward by combining HEER with three-dimensional (3D) reconstruction tomography technology. The validity and usage of the TDHEER method have been confirmed through simulation of the fully 3D diagnostic of static mesoscale sample. This paper focuses mainly on the experimental demonstration of the 3D high energy electron radiography. The feasibility of TDHEER is for the first time confirmed by the results achieved with different 3D reconstruction algorithms. The 3D reconstruction algorithms, analytical algorithm-filtered back projection (FBP), iterative algorithms-algebraic reconstruction technique (ART), and simultaneous algebraic reconstruction technique (SART) are used here. In this experiment, the less projected data are used, so it takes the less time to obtain two-dimensional (2D) HEER images and the reconstruction. In order to spend the time as little as possible and obtain the satisfactory quality of reconstruction result, there are three groups of projected image sets, 180, 36 and 18, acquired in our experiment. When all three algorithms are adopted in 180 projected images, the reconstructed images show that all three algorithms FBP, ART and SART are feasible for TDHEER. The different reconstructed slice images of the sample in X-, Y-, and Z-direction clearly show the detailed structure of the sample. The images reconstructed by ART and SART algorithm are equivalent. Comparing with ART and SART, the reconstruction results by FBP can show more details, but there are some artifacts. Because the 36 2D HEER images fail to satisfy the Nyquist sampling theory, the analytic algorithm FBP is not used. Taking the result of FBP reconstructed by 180 images as a standard reference to compare the result of ART with the results of SART, the images reconstructed by the SART algorithm are closer to the original images. Testing 18 images, the results of the ART and SART both have lots of artifacts but the SART algorithm spends less time in reconstruction. As fewer projected images are used, more artifacts are found in the reconstructed images. Therefore, it is advantageous to combine the SART algorithm with 36 HEER projected images, which obtains high-quality reconstruction images and spends less time. The feasibility of TDHEER is confirmed experimentally for the first time and all three dimensions of the sample structures are obtained. Of the three different 3D reconstruction algorithms, the SART algorithm is the most suitable for reconstructing the few-view images. The TDHEER technology will extend HEER’s application fields, especially for mesoscale sciences.