We present a new perspective on the Schottky problem that links numerical computing with tropical geometry. The task is to decide whether a symmetric matrix defines a Jacobian, and, if so, to compute the curve and its canonical embedding. We offer solutions and their implementations in genus four, both classically and tropically. The locus of cographic matroids arises from tropicalizing the Schottky-Igusa modular form.arXiv:1707.08520v2 [math.AG] 30 Sep 2018 the latter, τ already passed that test, and we compute a curve whose Jacobian is given by τ . The recovery problem also makes sense for g = 3, both classically [4] and tropically [1, §7].This paper is organized as follows. In Section 2 we tackle the classical Schottky problem as a task in numerical algebraic geometry [6,14,24]. For g = 4, we utilize the software abelfunctions [25] to test whether the Schottky-Igusa modular form vanishes. In the affirmative case, we use a numerical version of Kempf's method [17] to compute a canonical embedding into P 3 . Our main results in Section 3 are Algorithms 3.3 and 3.5. Based on the work in [7,8,27,28], these furnish a computational solution to the tropical Schottky problem. Key ingredients are cographic matroids and the f-vectors of Voronoi polytopes.Section 4 links the classical and tropical Schottky scenarios. Theorem 4.2 expresses the edge lengths of a metric graph in terms of tropical theta constants, and Theorem 4.9 explains what happens to the Schottky-Igusa modular form in the tropical limit. We found it especially gratifying to discover how the cographic locus is encoded in the classical theory.The software we describe in this paper is made available at the supplementary website http://eecs.berkeley.edu/~chualynn/schottky (2) This contains several pieces of code for the tropical Schottky problem, as well as a more coherent Sage program for the classical Schottky problem that makes calls to abelfunctions.