Abstract. We consider analysis of functional data which are only partially observed. Often in such cases, the observed fragments of curves are supported on quite different intervals, in which case standard methods of analysis cannot be used. We propose new approaches to analysing fragments of curves observed on different intervals. The techniques we suggest involve discretising the observed fragments, and then extending them outside the interval where they were observed. Using the same approach we can construct estimators of the mean and covariance functions, and, for example, deal with functional linear regression. This is joint work with Peter Hall.