Magnetotelluric measurements were performed simultaneously at two sites 4.8 km apart near Hollister, California. SQUID magnetometers were used to measure fluctuations in two orthogonal horizontal components of the magnetic field. The data obtained at each site were analyzed using the magnetic fields at the other site as a remote reference. In this technique, one multiplies the equations relating the Fourier components of the electric and magnetic fields by a component of magnetic field from the remote reference. By averaging the various crossproducts, estimates of the impedance tensor not biased by noise are obtained, provided there are no correlations between the noises in the remote channels and noises in the local channels. For some data, conventional methods of analysis yielded estimates of apparent resistivities that were biased by noise by as much as two orders of magnitude. Nevertheless, estimates of the apparent resistivity obtained from these same data, using the remote reference technique, were consistent with apparent resistivities calculated from relatively noise-free data at adjacent periods. The estimated standard deviation for periods shorter than 3 sec was less than 5 percent, and for 87 percent of the data, was less than 2 percent. Where data bands overlapped between periods of 0.33 sec and 1 sec, the average discrepancy between the apparent resistivities was 1.8 percent.