Fluorescence from endogenous molecules and exogenous contrast agents can provide morphological, spectral, and lifetime contrast that indicates disease state in epithelial tissues. Recently, nonlinear microscopy has emerged as a potential tool for the early detection, case-finding, and monitoring of epithelial cancers because it permits non-invasive, three-dimensional fluorescence imaging of subcellular features hundreds of microns deep. This dissertation explores the use of nonlinear microscopy for cancer diagnostics on two fronts: (1) we examine the fundamental limitations governing the maximum nonlinear imaging depth in epithelial tissues, and (2) we investigate the use of a new class of nonlinear contrast agent---plasmonic gold nanoparticles---for molecularly specific imaging of cancer cells. We built and optimized a nonlinear microscope for deep tissue imaging, and studied the image contrast as a function of imaging depth in ex-vivo human biopsies and tissue phantoms. With this system we demonstrated imaging down to 370 [mu]m deep in a human biopsy, which is significantly deeper than imaging depths achieved in comparable studies. We found that the large scattering coefficient and homogenous fluorophore distribution typical of epithelial tissues limit the maximum imaging depth to 3-5 mean free scattering lengths deep in conventional nonlinear microscopy. Beyond this imaging depth, the increasing contribution of out-of-focus emission limits the contrast to insufficient levels for diagnostic imaging. We support these observations with time-dependent Monte Carlo simulations. We exploited the intense interaction of gold nanoparticles with light, enhanced by surface plasmon resonance effects, to create extremely bright nonlinear contrast agents. These contrast agents proved to be several orders of magnitude brighter than the brightest organic fluorophores and at least one order of magnitude brighter than quantum dots. We targeted gold nanoparticles to a biomarker for carcinogenesis and demonstrated molecularly specific imaging of cancer cells. We demonstrated that unlike emission from traditional bandgap fluorophores, nonlinear luminescence from gold nanoparticles was weakly dependent on excitation pulse length for short pulse durations. This finding supports the hypothesis that nonlinear excitation in plasmonic nanoparticles involves sequential rather than simultaneous absorption of excitation photons. The remarkable brightness of gold nanoparticles makes them an attractive contrast agent for nonlinear diagnostics. / text