We study the problem of order optimality of an upper bound for the best approximation in L-q(T) in terms of the lth-order modulus of smoothness (the modulus of continuity for l = 1): En-1(f)(q) <= C(l, p, q) (Sigma(nu=n+1)nu(q sigma-1) omega(q)(l) (f; pi/nu)(p))(1/q), n is an element of N, on the class M-p(T) of all functions f is an element of L-p(T) whose Fourier coefficients satisfy the conditions a(0)(f) = 0, a(n)(f) down arrow 0, and b(n)(f) down arrow 0 (n up arrow infinity ), where l is an element of N, 1 < p < q < infinity, l > sigma = 1/p - 1/q, and T=(-pi, pi]. For l = 1 and p >= 1, the bound was first established by P. L.Ul'yanov in the proof of the inequality of different metrics for moduli of continuity; for l > 1 and p >= 1, the proof of the bound remains valid in view of the L-p-analog of the Jackson-Stechkin inequality. Below, we formulate the main results of the paper. A function f is an element of M-p(T) belongs to L-q(T), where 1 < p < q < infinity, if and only if Sigma(infinity)(n=1)n(q sigma-1) omega(q)(l) (f; pi/n)(p) < infinity, and the following order equalities hold: (a) En-1(f)(q)+n(sigma)omega(l)(f; pi/n)(p) asymptotic to (Sigma(infinity)(nu=n+1) nu(q sigma-1) omega(q)(l) (f; pi/v)(p))(1/q), n is an element of N, (b) n(-(l-sigma)) (Sigma(n)(nu=1) nu(p(l-sigma)-1) E-nu-1(P)(f)(q))(1/p) asymptotic to (Sigma(v=n+1) nu(q sigma-1) omega(q)(l) (f;pi/nu)(p))(1/q), n is an element of N. In the lower bound in equality (a), the second term n(sigma)omega(l)(f; pi/n)(p) generally cannot be omitted. However, if the sequence {omega(l)(f; pi/n)(p)}(n=1)(infinity) or the sequence {En-1(f)(p)}(n=1)(infinity) satisfies Bari's (B-l((p)))-condition, which is equivalent to Stechkin's (S-l)-condition, then En-1(f)(q) asymptotic to (Sigma(infinity)(nu=n+1) nu(q sigma-1) omega(q)(l)(f;pi/nu)(p))(1/q), n is an element of N. The upper bound in equality (b), which holds for every function f is an element of L-p(T) if the series converges, is a strengthened version of the direct theorem. The order equality (b) shows that the strengthened version is order-optimal on the whole class M-p(T).