In my book, "Probability and Stochastics" by Cinlar, it's stated that for some measurable space $(E,\scr E)$, and fixed $x\in E$, the Dirac measure $\delta_x(A)=\left\{ \begin{array}{lcc} 1, x\in A \\ \\ 0, &\mbox{ otherwsie} \\ \\ \end{array} \right.$
is purely atomic (i.e. The set $D$ of atoms of $\delta_x$ is countable and $\delta_x(E\setminus D)=0$, where an atom is a singleton $A$ such that $\delta_x(A)>0$).
I guess $\delta_x$ would have only one atom, that being $\{x\}$, but how can you say that $\{x\}\in\scr E$ for any $\sigma$-algebra $\scr E$? What if $\scr E$ is the trivial $\sigma$-algebra $\{E,\varnothing\}$?