For a complex polynomial D(t) of even degree, one may define the continued fraction of D(t). This was found relevant already by Abel in 1826, and later by Chebyshev, concerning integration of (hyperelliptic) differentials; they realized that, contrary to the classical case of square roots of positive integers treated by Lagrange and Galois, we do not always have pre-periodicity of the partial quotients.In this paper we shall prove that, however, a correct analogue of Lagrange's theorem still exists in full generality: pre-periodicity of the degrees of the partial quotients always holds. Apparently, this fact was never noted before.This also yields a corresponding formula for the degrees of the convergents, for which we shall prove new bounds which are generally best possible (halving the known ones).We shall further study other aspects of the continued fraction, like the growth of the heights of partial quotients. Throughout, some striking phenomena appear, related to the geometry of (generalized) Hyperelliptic Jacobians. Another conclusion central in this paper concerns the poles of the convergents: there can be only finitely many rational ones which occur infinitely many times. (This is crucial for applications to a function field version of a question of McMullen.)Our methods rely, among other things, on linking Padé approximants and convergents with divisor relations in generalized Jacobians; this shall allow an application of a version for algebraic groups, proved in this paper, of the Skolem-Mahler-Lech theorem.1 When λ = a/b is rational the procedure eventually terminates and corresponds to the Euclidean algorithm for a, b.The fraction pn/qn determines the polynomials pn, qn only up to a factor; usually here we implicitly mean that pn, qn are calculated formally from the an in the well-known natural way.3 This notion heavily depends on the ground field, but here we tacitly stick to C. 4 Already in the numerical case, Dirichet class-number formulae and other results indicate a strict connection of the topic with the suitable Picard groups.5 The case of polynomials over a finite field is, on the contrary, completely similar to the integer case. 6 A formal proof of this is the object of work in progress, but some detail appears already in [34], especially §2.2. It shall anyway clearly appear later that pellianity is indeed uncommon. 7 For reasons of space, we omit a discussion of this here, which is somewhat laborious, depending on Jacobians of dimension ≥ 3 containing a translate of an elliptic curve inside the set of sums of two points of the curve. To give a specific example, the polynomial D(t) = t 8 − t 7 − (3/4)t 6 + (7/2)t 5 − (21/4)t 4 + (7/2)t 3 − (3/4)t 2 − t + 1 yields infinitely many partial quotients of degrees 1 and 2, with the periodic pattern of degrees . See O. Merkert's thesis [22] for more. We plan to publish a detailed presentation in the future. 8 The methods of this paper should probably prove this for arbitrary elements of C(t, D(t)), though for simplicity we work only with the special ...