In this paper, an errors-in-variables (EIV) method is applied to the problem of model estimation for noise cancellation in transient electromagnetic mineral exploration. The algorithm exploits the non-stationary nature of the data. Alternative methods for noise cancellation in these systems rely on specific signal characteristics, and are thus less readily transferable to other applications. The proposed method produces a model that agrees well with those obtained by alternative methods and has similar noise cancellation performance. This is shown by performance comparisons on experimental data.