I have a quad (2 triangles) consisting of 4 (shared) vertexes which differ all 4 in x, y and z. On that quad I want to get a random position which is exactly ON the surface. The quad does not necessarily have to be a square, it can be anything with 4 vertexes. So I'll split the quad in two triangles...but then what? How can I ever get a random position on a non-square surface? I thought of calculating the surface area, then taking a random value between zero and the surface area value...but how would I translate that into a XYZ value? Thanks!

Use Barycentric coordinates.

To get a random point on a triangle ABC, generate the random values x (range [0,1]) and y (range [0,1-x]). Your random point is then A*x + B*y + C*(1-x-y).

Ah thanks!

One question though...

AB and C are vertexes, what do you mean when you say A * x? Most pages seem to explain the other way around, where you have a point and check if it's on the triangle...I haven't done much English written mathematics yet =/

A * x is a vector-scalar multiplication. A is a vector (multiple numbers, three in this case), x is a scalar (a single number).

The result is the component-wise multiplication of every number in A with x, i.e, B = A * x means that:
B.x = A.x * x
B.y = A.y * x
B.z = A.z * x

Thanks! It works xD

One thing though. I've been thinking about this in a 2D presentation to get a random point...

Imagine a triangle (2D) with one corner 90 degree, the other two 45 degree. Now to get a random position, you take a random Y value. Lets say the height and width of the triangle is 1 by 1. Y should be between 0 and 1...lets say it's 0.8. Now if you want to get the X value, you need to take the Y value into your calculations. Since it's not a square, the X value can now only be from 0 to 0.2! Since where 0.8Y is on the diagonal part, X is 0.2 and is then our maximal value.

This means the position of the point is never random. It has a random chance in height, but a much higher change to be closer to zero on the X axis.

Since I saw the Y value in your calculations used range(0, 1 - x), and I noticed that it's not 100% random, I thought the same might apply to your formula...

As solution I thought of this: every triangle can, fit twice in some sort of square (a rhomb, often). So if I take such a square, calculate a random point on the square, then whenever it's not on the triangle, I simply swap X and Y from direction (x = max_x - x)...would this work do you think?

Thanks!