The field of information theory provides a natural framework for the principled analysis of neural spike trains - the computational currency of the brain. Shannon information, in particular, has received the majority of attention. However, alternative measures of information such as Fisher information, derived from likelihood functions, may offer new insights into the form of information carried by cortical ensembles. Analysis utilizing Fisher information has typically concentrated on theoretical modeling, with less emphasis on applications to empirically recorded spike trains. Likelihood approaches are often directed at the topology of probability densities, which is often ignored by Shannon approaches to the measure of information. Our past neurophysiological studies have focused on single-unit neural coding of spatial hearing in auditory cortex. I will discuss the use of likelihood approaches to examine the impact of coding multiple parameter dimensions in the response of ensembles of cortical neurons. These approaches include asymptotic approximations to likelihood functions using simple Laplace and Saddlepoint approximations, and marginalizing neural likelihood functions to eliminate so-called nuisance parameters. We are ultimately interested in using these models to help explain observed patterns of psychophysical performance.