Automatic inference of the semantics of an image is still a highly challenging research problem in the computer vision area. It is concerned with applying computational and mathematical techniques, attempting to figure out the semantic meaning of the image content. In the Medical Imaging domain, this problem is even more complicated because of the overwhelming amount of prior medical knowledge that a Physician requires to cope with the variations of what is considered as the prototypical disease. This thesis addresses the main problems associated to the microscopic cyto and histo pathological image semantic analysis, including standardization of color and intensity characteristics, development of visual content representation methods that taken advantage of the particular characteristics of these images, and appropriate use of conventional learning models that allowed distinguishing different biological concepts in the images. This document presents the design, implementation and evaluation of strategies for reaching proper levels of semantic image interpretation, applied to two important microscopic applications: analysis and interpretation of cytological images for quantification of malarial infected erythrocytes, and analysis of micro-structural tissue components for automatic semantic annotation of histopathological images of skin biopsies diagnosed with basal-cell carcinoma. Obtained results outperform what has been so far reported in the literature for both applications, demonstrating the effectiveness and versatility of the proposed strategies. Main ideas and techniques developed in this work were also applied to the analysis and interpretation of other biomedical images as brain volumes and mammography. Results in these applications are included as document annex.