In one previous paper, we proposed a new model of error-detecting codes base on quasigroups on the following way. Each input block ai a2 . . . an is extended to a blockl a1a2 . . . anb1b2 . . . bn where bi = ai * ari i * ari 2 * ari k - I ' i E {1, 2, . . . , n}, * is a quasil group operation and r j = { j, d .i ,,::; n . We have already derived approximate formula! J mo n, J > n for the probability of undetected errors when quasigroups of order 4 are used for coding ancl k = 2. In this paper, we derive approximate formula for the probability of undetected error when also quasigroups of order 4 are used for coding, but k = 3. We find the optimal blockl length such that the probability of undetected errors is smaller than some previously given valu r=: and give classification of quasigroups of order 4 according to goodness for the code when k = 3.I Also, we compare these two considered codes and conclude that the best set of quasigroups forl coding for both codes contains only linear fractal quasigroups and the code with k = 3 give much smaller probability of undetected errors. At the end, we compare the code considered inl this paper with well-known error-detecting codes: CRC, Hamming and Reed-Muller.