Abstract: A reconstruction technique of bioluminescence sources in natural waters from {\it in situ} irradiance data is presented. The inverse problem is formulated as a nonlinear constrained optimization problem, assuming that the bioluminescence unknown profile can be represented by a sum of distributed gaussian sources. The objective function is defined as the square Euclidean norm of the difference vector between experimental and computed data. The associated direct problem is tackled with the Hydrolight 3.0 code. This model solves numerically the time-independent, one-dimensional radiative transfer equation in natural water bodies using the invariant imbedding theory. The proposed inversion technique was tested with noise-corrupted synthetic data, and yielded good numerical results. The influence of the number of gaussian sources and their standard deviations in the estimation is analysed.