When crowds gather to make political statements, it matters how many people turn out. Crowd size matters to organizers, who invariably say they made their point. It matters to police departments, who insist they fielded the right number of officers. It matters to the media, who often claim they've reported the facts. And it matters to elected officials, who often like to act as if the whole thing never happened.

All of which is why February's antiwar demonstrations came under such close scrutiny. Consider what happened in San Francisco. Organizers, whose tallies usually are made by observers at ground level, initially estimated turnout at 250,000 (later revised to 200,000). Police, who tend to eyeball their figures from the air based on known counts such as stadium capacities, also claimed 200,000. But an aerial photo survey commissioned by the San Francisco Chronicle calculated just 65,000 at the peak time of 1:45 pm.

Photographic analysis of the kind undertaken by the Chronicle can provide a verifiable estimate, but it must be done correctly. Here's how:

· Fly over the crowd at peak times using a fixed-wing aircraft. (Shaky helicopter platforms blur photos, increasing the effort required to analyze them.) Altitude should be 2,000 feet or less.

· Photograph the area in strips using a digital camera, with 60 percent overlap between successive pictures to allow stereoscopic viewing (helpful for making ambiguous pictures clearer). Image resolution should be about 1 foot per pixel.

· Load the photos in an image-processing program and co-register them with a 1-meter-resolution US Geological Survey orthophotomap – a perspective-corrected collage of aerial shots of the area with a uniform scale.

· Superimpose a grid on the image and classify its units by the apparent density of people per unit.

· Place a cross or dot on each individual head or shadow point.

· Count or, if necessary, estimate the number of people in each grid unit, then tally the numbers.

· Calculate error – basically the number of grid units divided by the degree of uncertainty about how many people they contain.

By these criteria, the Chronicle's survey fell short. For one thing, it was based on analog photos, which lose resolution when digitized. It also failed to co-register the images with an orthophotomap, and the grid unit size was larger than 1 meter, too big for careful counting.

Still, the newspaper's effort was a step in the right direction. Where political interests are at stake, the job of counting masses of people falls naturally to the media. It doesn't cost much – $2,000 to $5,000 per event, depending on location and crowd size. A joint effort by news outlets and independent supervisors, especially in politically sensitive cases, would bring crowd counts out of the murky realm of guesswork and into the bright light of verifiable fact.