Summary: Let \({\mathfrak S}\) be a finite subset of \(\mathbb{Z}^2\). A walk on the slit plane with steps in \({\mathfrak S}\) is a sequence \((0,0)= w_0,w_1,\dots, w_n\) of points of \(\mathbb{Z}^2\) such that \(w_{i+ 1}- w_i\) belongs to \({\mathfrak S}\) for all \(i\), and none of the points \(w_i\), \(i\geq 1\), lie on the half-line \({\mathcal H}= \{(k, 0): k\leq 0\}\). In a recent paper [Walks on the slit plane, preprint 2000 arXiv: match.CO/0012230] G. Schaeffer and the author computed the length generating function \(S(t)\) of walks on the slit plane for several sets \({\mathfrak S}\). All the generating functions thus obtained turned out to be algebraic: for instance, on the ordinary square lattice, \(S(t)= ((1+ \sqrt{1+ 4t})^{1/2}(1+ \sqrt{1- 4t})^{1/2})/2(1- 4t)^{3/4}\). The combinatorial reasons for this algebraicity remain obscure. In this paper, we present two new approaches for solving slit plane models. One of them simplifies and extends the functional equation approach of the original paper. The other one is inspired by an argument of Lawler; it is more combinatorial, and explains the algebraicity of the produt of three series related to the model. It can also be seen as an extension of the classical cycle lemma. Both methods work for any set of steps \({\mathfrak S}\). We exhibit a large family of sets \({\mathfrak S}\) for which the generating function of walks on the slit plane is algebraic, and another family for which it is neither algebraic, nor even D-finite. These examples give a hint at where the border between algebraicity and transcendence lies, and call for a complete classification of the sets \({\mathfrak S}\).