Most quantitative measures of chaos (e.g., fractal dimensions or Lyapunov exponents) rely on constructing an approximation of the natural measure on a strange attractor, which requires observing the system for at least a few hundreds of cycles at fixed control parameters. Thus, it is extremely difficult to assess deterministic chaos in a real system that experiences parameter drifts on a time scale comparable to the mean dynamical period. A natural question then is: can we infer the existence of an underlying chaotic dynamics from a very short, nonstationary, time series?We present an experimental case in which this question can be answered positively. By applying topological tools to a burst of irregular behavior recorded in a triply resonant optical parametric oscillator subject to thermal effects, we have extracted a clearcut signature of deterministic chaos from an extremely short time series segment of only 9 cycles. Indeed, this segment shadows an unstable periodic orbit whose knot type can only occur in a chaotic system. Moreover, this topological approach provides us with quantitative estimates of chaos, as a lower bound on the topological entropy of the system can be determined from the knot structure. Two positive‐entropy periodic orbits are detected in a time series of about 40 cycles, suggesting that the presence of such orbits in a time series is common. Thus, nonstationarity is not necessarily an obstacle to the characterization of chaos. © 2004 American Institute of Physics