In any real-life identification problems, only a finite number of data points are available. On the other hand, almost all results in stochastic identification pertain to asymptotic properties, that is they tell us what happens when the number of data points tend to infinity. In this paper, we consider the problem of assessing the quality of non-asymptotic estimates obtained using least squares identification methods. The type of results needed in order to be useful for computing the quality of non-asymptotic estimates are first discussed. It turns out that the nature of non-asymptotic results has to be different from that of asymptotic results, since in finite time certain issues show up that disappear in the limit because of stochastic convergence. Then, we develop a method for the assessment of the estimate quality based on differences between partial estimates. If the partial estimate differences are within a small region around zero then, as it is intuitive, the estimate quality is good. On the other hand, we will have low confidence in the estimate if the differences between partial estimates are spread all over the place. The method is illustrated through a very simple example able to point out its main aspects in a clear-cut way.