What is the ideal texture resolutions for clothing/skins?

I'm trying to create clothes/skins. The canvas size that I work on in PS is 1024 x 1024 pixels and I don't change anything when I upload. When I try the textures on however, they seem a little blurry and not as sharp as I would like. I try to "increase sharpness" which improves this problem a little but not much. Is there anything I could do to make the texture the best quality it can be for upload to SL?

Thanks, all feedback is appreciated =]

Upload at 512x512 because they will be rescaled to that size anyway when the clothing layers are "baked".

To expand on what Argent said, here's how it works. When you put an article of clothing on your avatar, the texture for that article is downloaded to your local viewer. The viewer then composites that texture, along with every other texture your avatar is wearing, into one master set of three 512x512 images (head, upper body, and lower body), called an "outfit". That outfit is then uploaded to the sim server, so that everyone else in visual range can see your avatar fully "dressed".

It makes no difference what sizes the source textures are. That new shirt you just made (or bought) could be anything from 8x8 to 1024x1024, and it's going still going to get resized to 512x512 when the outfit is composited, no matter what.

The best thing to do is to work at a larger size in Photoshop, or whatever your raster editor of choice happens to be, and then downsize to 512x512 as a final step before uploading to SL. If you upload a 1024x1024, all you do is make the compositing process take a little longer, since a 1024x1024 will take longer to download than a 512x512. Also, you sacrifice some control, since the SL compositor will use just one downsizing algorithm, as opposed to the choice of algorithms Photoshop offers. For best visual clarity, not every image should be downsized exactly the same way.

To go even further in depth on what Chosen has said, all the Creative Suite releases of Photoshop (8,9,10) have specific algorithms for enlargement (Bicubic Smoother) and reduction (Bicubic Sharper). Previous releases of PS had to rely on bicubic interpolation and manual adjustments with "Unsharp Mask", or using nearest neighbor interpolation, which preserves hard edges but can cut small strips of pixels out of an image during reduction. The missing pixel strips usually occur within the image and not on the edge. The aspect ratio is also altered to compensate for this depending upon how information is rounded off. Fortunately, clipped pixel strips and aspect ratios are not a problem with a 1024 to 512 reduction. Photoshop's dialog box should calculate and confirm this information before reduction is executed.

My advice to the OP is to use "Bicubic Sharper" (if you can) for the needs discussed here, since it is the most elegant, efficient, and safest method. If that's not available to you, then try "Nearest Neighbor", since you should be safe from the pitfalls described above. Nearest neighbor interpolation actually looks sharper than "Bicubic Sharper" in many cases. That may fly in the face of logic, but I speak from experience. It's the quickest way to deal with image blurring during reduction with earlier versions of PS in special cases such as this. Most other raster image editors will also have nearest neighbor interpolation as an option too. In many other cases, when using an older version of Photoshop, some form of unsharp masking performed before reduction, using bicubic interpolation, will yield the closest and safest result to "Bicubic Sharper" because you won't risk clipping pixel strips out of your image.