Much effort has been undertaken over the past two decades to elucidate the dynamical properties of chaotic processes from time series measurements. One example is the estimation of Lyapunov exponents from time delay embeddings. However, the results are subject to large uncertainties. This talk will argue that there are no obvious probability models for errors that arise from artifacts of the embedding function or from poor approximations of the natural measure of the underlying attractor. Consequently, the construction of confidence intervals for dynamical quantities like Lyapunov exponents does not seem feasible. Nevertheless, it may still be possible to get a reasonable idea of the size of the uncertainties associated with their computation from nonlinear time series.