The resolution of the image of the blueprints is more important than the size, and all other things being equal, in choosing between to sets of blueprints of the same model, I'll go with the set with the higher resolution.
Here is a description of my workflow in processing the image for use in Blender, although the first bits of this are done in image manipulation software (My choice is GIMP, but I expect others would work as well) do not involve blender at all. For the sake of the illustration, I'll assume that the object to be modeled is a 14' 6" high (top or rail to top of car) boxcar, 40' 6" long, and 9' 6" wide.
1) I separate each of the various views of the blueprint or plan into its own separate image, so I'll have an end view, a side view, a top view, and in the case of drawings from the model railroad hobbyist press, sometimes one or more cross sectional views, too.
Second, I compare common dimensions between pairs of the images resulting from step one, so that I'll check the maximum height dimension (in pixels) of the end view, and compare it to the maximum height dimension of the side view. If necessary, I scale the image with the larger height dimension (in pixels) down by the proportion that makes the two dimensions equal to the smaller. I'll check many (if not all) dimensions between pairs of images in this manner, for example, the width of the car, between end and top views, and the length of the car in top and side views, &c. When I'm done with this step, all major dimensions will be the same number of pixels in whatever view one chooses to look at them.
3) I take the largest image (usually the side view), and sometimes increase the size of the "canvas" to human friendly numbers. For example, if the image is 732 pixels high, and 3127 pixels long, I'll likely place it on a canvas 800 x 3200 pixels. Then using a feature in GIMP (and I don't know what other software has this feature) I'll center the image on the "canvas". I leave the border around the center image blank.
4) I repeat step three for all of the other images, using the same size canvas, leaving the borders around the image blank, and centering the image on the canvas. So, all my reference images have are the same dimensions in pixels, even though the content of interest on, for example, an end view, may be much smaller.
5) Then I measure the length of one of the dimensions of the car in pixels, and use that to determine the number of pixels per foot. Let's say the 14' 6" height of the box car is 712 pixels. From this, it can be calculated that the number of pixels per foot is about 49, so the 800 pixel image canvas size will be a scale 16.32 feet, and applying the same process to the length of the image canvas (3200 pixels), the scale length of the image is calculated to be 65.30.
6) Here is where I start Blender. I start with a default scene, leave the camera and light in layer 1, and move the cefault cube to layer 3. In layer 2, I create a plane, and scale the plane so that the dimensions are 16.32 feet by 65.30. I duplicate the plane, and rotate the duplicate so that the shorter side is vertical, and the longer side is horizontal.
7) I switch to the UV editing setup, and load (in this instance) the side view of the car as the image. I UV unwrap the plane, and assign the image of the car side to the plane.
8) I repeat steps 6 and 7 for each of the other images, orienting the planes appropriately for the end, top, and any other views that might be available.
9) Leaving the UV editing window, I go to layer 2, and make sure that all of the planes are correctly oriented, so that as one looks to the end view, one sees the end of the car. I also make sure the views are all aligned, so that the top of the car in each image matches the top of the car in every other image (where it shows).
10) I go to layer 3, and either start using the default cube to make my model, or delete it and start from scratch. When I need to see a reference image, I simply activate layer 2 in the object window, if the reference images are getting in the way, I can either de-activate layer 2, or hide the reference that's interfereing in the outliner window.
Hope this helps
ns
