Scientists from the Howard Hughes Medical Institute, National Institutes of Health, and Florida State University created what they believe is the highest resolution 3D optical microscope. By upgrading a PALM (photoactivated localization microscopy) device using interferometry techniques, they were able to bring out the third dimension.
From a statement by HHMI:

Hess [Harald F. Hess] and Janelia Farm colleague Eric Betzig invented the PALM microscope in 2005. Scarcely three years later, it was one of a handful of new methods of “super-resolution” microscopy that were honored by Nature Methods in January 2009 as the “Method of the Year” for the previous year.
PALM permits biologists to visualize cells with far more detail than conventional optical microscopes, which are inherently limited by the wavelength of light. To achieve this resolution, PALM uses fluorescent labels that can be turned on and off with a pulse of light. Cells whose proteins are tagged with these labels are imaged repeatedly with PALM, with only a tiny subset of the fluorescent molecules turned on in each image. By compiling many thousands of these images, PALM creates a complete picture of the structure under study, pinpointing each fluorescently tagged protein. As a result, researchers get a much clearer picture than the overlapping haze that results when all of the tagged proteins are lit up at the same time, as in traditional fluorescence microscopy.
Hess, who spent eight years working in the data storage and semiconductor industries, quickly focused his thinking on interferometry as a way to identify a protein’s precise depth within a biological sample, and September of 2006 proposed the idea of iPALM. “Interferometry is one of the more sensitive measurement techniques out there,” Hess said. “If you have bright enough light sources, you can measure ridiculously tiny displacements – way below the size of an atom.”
When he worked in the hard disk industry, Hess used interferometry to detect subtle convolutions on the surface of a hard drive disk. The approach, he said, involved bouncing light off the surface of the disk and comparing the returned light wave to a “reference wave,” which had been bounced off a mirror a known distance from the light source. “If light goes down and bounces off a surface, if that surface is a little bit higher or a little bit lower, that wave’s going to be coming at you a little bit later or a little bit sooner,” he explained. If the mirror and the experimental surface are the same distance from the light source, the waves, when added together, will cancel one another out. But tiny discrepancies in the two distances will shift the waves a measurable amount. “Depending on the amplitude of the summed waves,” he said, “you can determine the vertical position to within nanometers.”
No one had figured out how to apply the technique to biological samples, however. The primary challenge, Hess explained, was that in fluorescence microscopy, the key light waves travel from fluorescent tags within the sample itself, not from a readily manipulated laser. “It’s a whole new paradigm,” he said. “It isn’t like you can go in there and take a piece of the laser to make a reference beam.”
Hess and Janelia Farm colleague Gleb Shtengel saw a way around the problem: They decided to split each particle of light emitted from the fluorescent molecule in two. By splitting the photons, the researchers knew that each fluorescent photon would act as its own reference beam. They adapted the standard PALM microscope to collect this light both above and below the sample. Both of those beams of light travel to a custom-made beam-splitter, which divides the beam and sends it to three different cameras. A molecule’s depth within the sample determines how much light reaches each of the cameras. “We record an image triplet, and depending on how much appears in camera one, two and three, we can say ‘this was the height.’ This is by far the most sensitive way of measuring vertical height,” Hess said.
“iPALM needs only a modest amount of light to generate its sensitive measurements, and that’s important for biological imaging,” Hess says. Imaging techniques that demand more photons can force researchers to label the proteins they want to see with brighter dyes – which are often bulky and require harsh sample preparations that damage cells. Fluorescent probes such as those compatible with iPALM, on the other hand, can be genetically encoded so that they are manufactured by cells themselves. The power of these glowing markers was recognized with the 2008 Nobel Prize in Chemistry, which was awarded to the HHMI investigator Roger Y. Tsien, Osamu Shimomura, and Martin Chalfie for the discovery and development of the first such tool, green fluorescent protein.

View 3D video of integrin proteins as visualized with iPALM
iPALM Tutorial PDF
Press release: Super-Resolution Microscopy Takes on a Third Dimension
Image: Side: With PALM imaging, the two dimensional distribution of the labeled membrane proteins becomes much clearer. However, it is impossible to determine the vertical position of the fluorescent molecules in the flat image. Top: iPALM pinpoints the three-dimensional distribution of the fluorescently tagged membrane proteins. In this image, the vertical position has been color coded, with red molecules being the deepest and purple the highest. Cross-sections of small regions of the image are shown in the white boxes on the right, and reveal two layers of the labeled membrane proteins — at the top and bottom of the cell.