You can always use the microcanonical ensemble. In the Ising model, this amounts to fixing some (admissible) energy $E$ and considering the uniform measure on the set of all configurations with energy $E$.
Let me take as an example the simplest situation in which the probability of up an down spins are different: the case of independent spins $\sigma_1,\dots,\sigma_N$ in a field, that is, the Ising model with Hamiltonian $H(\sigma) = -h\sum_{i=1}^N \sigma_i$. The admissible energies in this case are those in the set $\mathcal{E} = \{-Nh, (-N+2)h,\dots, (N-2)h, Nh\}$.
Given some energy $E\in\mathcal{E}$, the corresponding microcanonical measure is then simply the uniform measure on the set
$$
\{\sigma\in\{-1,1\}^N\,:\, H(\sigma) = E\}.
$$
Of course, in this trivial case, this set can easily be described explicitly: if $E=nh$, then a configuration has energy $E$ if and only if there are $(N-n)/2$ spins equal to $+1$ and $(N+n)/2$ spins equal to $-1$.
Of course, the general scheme remains true if you include interactions between the spins.