Heatsink performance is measured in °C/W (or K/W - since we're dealing with temperature differences, there is no difference between Celsius and Kelvin scale here). We refer to this as thermal resistance (θ).

An example for what these values mean: if a thermal load of 20W is applied to a heatsink, and this causes the temperature of the heat source to raise by 10°C, the heatsink has a rating of of 10°C/20W = 0.5°C/W.

A θ value is valid only for a certain power load and a certain temperature range.

The thermal resistance of standard coolers for PC CPUs is usually not specified by the heatsink manufacturers, and if it is, it's often inaccurate or intentionally skewed for marketing purposes. You cannot judge heatsink performance by comparing θ specifications from different manufacturers.

The θ values specified by manufacturers specialized in heatsinks for industrial applications (especially large passive heatsinks) are usually more accurate, though.

Heatsink testing is not an easy task; many of the heatsink reviews found on the countless cooling-related sites on the net are not done properly and proper care must be taken to ascertain the validity of any empirical data provided.