The previous chapters considered empirical likelihood based on observations from the distribution (or distributions) of interest. This chapter considers empirical likelihood inference in some nonstandard sampling settings. In biased sampling, the data are sampled from a distribution different from the one we want to study. In censoring, some of the observations are not completely observed, but are known only to belong to a set. The prototypical example is the time until an event. For an event that has not happened by time t, the value is known only to be in (t,∞). Truncation is a more severe distortion than censoring. Where censoring replaces a data value by a subset, truncation deletes that value from the sample if it would have been in a certain range. Truncation is an extreme form of biased sampling where certain data values are unobservable.