We carried out an extensive paleointensity study of the 15.5±0.3 m.y. Miocene reversed‐to‐normal polarity transition recorded in lava flows from Steens Mountain (south central Oregon). One hundred eighty‐five samples from the collection whose paleodirectional study is reported by Mankinen et al. (this issue) were chosen for paleointensity investigations because of their low viscosity index, high Curie point and reversibility, or near reversibility, of the strong field magnetization curve versus temperature. Application of the Thellier stepwise double heating method was very successful, yielding 157 usable paleointensity estimates corresponding to 73 distinct lava flows. After grouping successive lava flows that did not differ significantly in direction and intensity, we obtained 51 distinguishable, complete field vectors of which 10 are reversed, 28 are transitional, and 13 are normal. The record is complex, quite unlike that predicted by simple flooding or standing nondipole field models. It begins with an estimated several thousand years of reversed polarity with an average intensity of 31.5±8.5 μT, about one third lower than the expected Miocene intensity. This difference is interpreted as a long‐term reduction of the dipole moment prior to the reversal. When site directions and intensities are considered, truly transitional directions and intensities appear almost at the same time at the beginning of the transition, and they disappear simultaneously at the end of the reversal. Large deviations in declination occur during this approximately 4500±1000 year transition period that are compatible with roughly similar average magnitudes of zonal and nonzonal field components at the site. The transitional intensity is generally low, with an average of 10.9±4.9 μT for directions more than 45° away from the dipole field and a minimum of about 5 μT. The root‐mean‐square of the three field components X, Y, and Z are of the same order of magnitude for the transitional field and the historical nondipole field at the site latitude. However, a field intensity increase to pretransitional values occurs when the field temporarily reaches normal directions, which suggests that dipolar structure could have been briefly regenerated during the transition in an aborted attempt to reestablish a stationary field. Changes in the field vector are progressive but jerky, with at least two, and possibly three, large swings at astonishingly high rates. Each of those transitional geomagnetic impulses occurs when the field intensity is low (less than 10 μT) and is followed by an interval of directional stasis during which the magnitude of the field increases greatly. For the best documented geomagnetic impulse the rapid directional change corresponds to a vectorial intensity change of 6700±2700 nT yr⁻¹, which is about 15–50 times larger than the maximum rate of change of the nondipole field observed during the last centuries. The occurrence of geomagnetic impulses seems to support reversal models assuming an increase in the level of turbulence within the liquid core during transitions. The record closes with an estimated several thousand years of normal polarity with an average intensity of 46.7±20.1 μT, agreeing with the expected Miocene value. However, the occurrence of rather large and apparently rapid intensity fluctuations accompanied by little change in direction suggests that the newly reestablished dipole was still somewhat unstable.