The curl of a vector field [itex]\vec{F}(x,y,z)[/itex] can be written as
[tex]\mbox{curl}\vec{F} =\left(\frac{\partial F_z}{\partial y}-\frac{\partial F_y}{\partial z},\frac{\partial F_x}{\partial z}-\frac{\partial F_z}{\partial x}, \frac{\partial F_y}{\partial x}-\frac{\partial F_x}{\partial y} \right)[/tex]
But also in the more illuminating way:
[tex]\mbox{curl}\vec{F}=\lim_{R\rightarrow 0}\frac{\oint_{C_R}\vec{F}\cdot d\vec{r}}{2\pi R}[/tex]
where [itex]C_R[/itex] is a circle of radius R. That is to say, the curl of F at (x,y,z) is the path integral of F around a tiny circle centered on (x,y,z) [divided by the length of its its circumference].
With this definition of curl, you can see intuitively why the vector field in Meir Achuz's post is not curlless, and more generally, why the curl will have maximum value at a given point when the vector field is circulating around that point, but that it is not necessarily zero otherwise. It is only in this broad sense that we mean that the curl is a measure of the circulation of a vector field.