Orbit determination is possible for a chaotic orbit of a dynamical system, given a finite set of observations, provided the initial conditions are at the central time. We test both the convergence of the orbit determination procedure and the behavior of the uncertainties as a function of the maximum number n of map iterations observed; this by using a simple discrete model, namely the standard map. Two problems appear: first, the orbit determination is made impossible by numerical instability beyond a computability horizon, which can be approximately predicted by a simple formula containing the Lyapounov time and the relative roundoff error. Second, the uncertainty of the results is sharply increased if a dynamical parameter (contained in the standard map formula) is added to the initial conditions as parameter to be estimated. In particular the uncertainty of the dynamical parameter, and of at least one of the initial conditions, decreases like n^a with a<0 but not large (of the order of unity). If only the initial conditions are estimated, their uncertainty decreases exponentially with n, thus it becomes very small. All these phenomena occur when the chosen initial conditions belong to a chaotic orbit (as shown by one of the well known Lyapounov indicators). If they belong to a non-chaotic orbit the computational horizon is much larger, if it exists at all, and the decrease of the uncertainty appears to be polynomial in all parameters, like n^a with a approximately 1/2; the difference between the case with and without dynamical parameter estimated disappears. These phenomena, which we can investigate in a simple model, have significant implications in practical problems of orbit determination involving chatic phenomena, such as the chaotic rotation state of a celestial body and a chaotic orbit of a planet-crossing asteroid undergoing many close approaches.