The problem of computing orbital elements of a binary from a set of observed positions is formally analogous to the case of orbits in the planetary system, yet in practice there is little resemblance between the methods. The motion of the earth is insignificant (or easily allowed for in the spectroscopic and photometric cases) which greatly simplifies the formalism. On the other hand, the measuring errors are much larger, compared with the quantities to be determined, in the binary-star case, so that an orbit needs to be based on a much larger number of observations.