In the Sequential Migration Aided Reflection Tomography method (Grau and Lailly, 1993), travel times used by reflection tomography are computed by tracing rays which propagate with the migration velocity and reflect from reflectors picked on migrated images. Because of limits of migration resolution, this picking involves inaccuracies, to which computed travel times are unfortunately very sensitive. The objective of this paper is to predict a priori the confidence we can have in emergence data, i.e., emergence point location and travel time, from the statistical information that describes the uncertainties of the reflectors. (These reflectors can be obtained by picking on migrated images as explained above or by any other method.) The proposed method relies on a linearization of each step of the ray computation, allowing one to deduce, from the statistical properties of reflector fluctuations, the statistical properties of ray-tracing outputs. The computed confidences and correlations give access to a more realistic analysis of emergence data. Moreover, they can be used as inputs for reflection tomography to compute models that match travel times according to the confidence we have in the reflector. Applications on real data show that the uncertainties are generally large and, what is much more interesting, strongly varying from one ray to another. Taking them into account is therefore very important for both a better understanding of the kinematic information in the data and the computation of a model that matches these travel times. ¿ American Geophysical Union 1995