This talk will describe bridges between our empirical understanding of neural information coding in biological systems and our theoretical understanding of information science. The first part of the talk starts with biology, working towards information theory and reviewing our current understanding of how information is coded in neural spike trains. There has been a change from a view of neurons as highly noisy components which convey information primarily through mean firing rate, to a more recent view in which neurons are understood as highly tuned and efficient information encoders which make use of precise spike timing in their encodings. An understanding of the neural codes can be gained through taking the "organism's view" and using neural output to make inferences about the stimuli which caused those outputs. We will review methods for reconstructing neural stimuli from spike encodings. These methods have been used on a variety of neural systems, such as mechanosensory cells in crickets and ganglial cells in amphibians, and reveal that the coding efficiency of these systems can be quite high. (e.g. experiments on sacculus cells in salamanders put a lower bound on coding efficency in the 50-60% range.) We then start from information theory and work towards biology, suggesting some basic generic optimality assumptions which seem plausible for highly evolved sensory information systems. Transmission of spikes is metabolically expensive, and increased rates of information transmission require greater expenditure of energy. Two plausible assumptions are that: (1) at any given rate, efficient encoding should minimize the energy dissipation at that rate, and (2) in any behavioral regime, the rate chosen by the organism must be understood in the broader context of tradeoffs between metabolic costs and potential gains through improved behavior. We will close by describing some consequences of these assumptions, such as the required statistical structure of efficient codes, and discuss current work to tie those predictions back to empirical data from biological experiments.