This paper reviews software (soft) sensors applied to bioreactor monitoring and control. Soft sensors are used to infer on-line variables for which hardware sensors are not available. Three main classes of software sensors are reported in the literature: 1 - sensors based on mechanistic models obtained from first principle equations; 2 - sensors based on black box empirical models such as neural networks or nonlinear polynomial models; 3 - sensors based on a combination of the first two classes. Examples of these three categories are given and comparisons are drawn between the different methodologies. Also, the suitability of the different sensors for each application is established.