Torsion points on J0(N) and Galois representations

Suppose that N is a prime number greater than 19 and that P is a point on the modular curve X-0(N) whose image in J(0)(N) (under the standard embedding iota: X-0(N) --> J(0)(N)) has finite order. In [2], Coleman-Kaskel-Ribet conjecture that either P is a hyperelliptic branch point of X-0(N) (so that N is an element of {23, 29, 31, 41, 47, 59, 71}) or else that iota(P) lies in the cuspidal subgroup C of J(0)(N). That article suggests a strategy for the proof: assuming that P is not a hyperelliptic branch point of X-0(N), one should show for each prime number a that the l-primary part of iota(P) lies in C. In [2], the strategy is implemented under a variety of hypotheses but little is proved for the primes l = 2 and l = 3. Here I prove the desired statement for l = 2 whenever N is prime to the discriminant of the ring End J(0)(N). This supplementary hypothesis, while annoying, seems to be a mild one; according to W. A. Stein of Berkeley, California, in the range N < 5021, it is false only in case N = 389.