In using the present primary standard of light-which is a full radiator held at the freezing point of platinum-to reproduce the basic unit of light, the candela, it has proved both difficult and time-consuming to achieve the required accuracy. Similar troubles with earlier material standards prompted various proposals for trying to place photometry on a radiometric basis, not requiring the use of a material standard of light, but none of these came to fruition. In the present paper a new method is put forward, which is claimed to be both sound and effective, for the calibration of tungsten filament secondary standard lamps, and which does not require actual use of the primary standard. It consists in first passing the radiation from the lamp through a filter whose spectral transmission curve is a proportional `copy' of the V$_\lambda$ function (the relative spectral response of the standard eye), and then measuring the resultant total radiation intensity with a calibrated radiation thermopile. After allowance for the scale factor relating the filter curve to the actual values of V$_\lambda$, which factor can be found from spectrophotometric measurements, this radiation intensity clearly is an absolute measure of the luminous intensity of the lamp, needing only to be multiplied by some fixed factor for conversion from the power scale to the light scale. This factor is, in fact, the ratio between the K$_\lambda$ curve, representing the absolute luminous efficiency of radiation (in lumens per watt, for example), and the V$_\lambda$ curve which represents the same function on a relative scale and normalized to unity at its peak value. This ratio can be computed, although not without some uncertainty, from the known physical properties of the primary standard; or it may be found experimentally by making observations on either the primary standard or existing tungsten filament standard lamps with the new method. It is an advantage that if, as at the National Physical Laboratory, the thermopile is calibrated in terms of the watt by electrical means instead of by exposure to a standard full radiator, any small future adjustments in the values of the radiation constants or of the temperature assigned as the freezing point of platinum, will not affect the results. Preliminary experiments with the method are described. They suggest that the precision may significantly surpass what has so far been obtained in the use of the primary standard of light. As applied to existing N.P.L. standard lamps, the method has given the result K$_\lambda$/V$_\lambda$ = 685 lumens per watt, as compared with a computed ratio of about 680. The paper concludes with a discussion of the implications of any possible future changes in V$_\lambda$, and of such possible changes in the definition of the primary standard or the unit of light as the method itself seems to suggest.