Abstract: A method and an arrangement for quality estimation when detecting an information frame, for instance a speech frame transmitted in a radio communication system in accordance with time division multiple access (TDMA), although the method can also be applied with frequency divided radio systems. There is used the soft information available in the radio receiver which is intended to detect the speech frame and process this signal information in accordance with those principles applicable to known neural nets. Prior to use, the neural net is aligned to the radio communication system concerned. The invention enables better information relating to possible error in the speech frame to be delivered to error correcting units in the radio receiver.