Two methods of decoding instruction were compared. Participants were kindergartners who knew letter sounds but could not decode nonwords, M = 5.6 years. The segmented phonation treatment taught students to convert graphemes to phonemes by breaking the speech stream ("sss -- aaa -- nnn") before blending. The connected phonation treatment taught students to pronounce phonemes without breaking the speech stream ("sssaaannn") before blending. The CVC nonwords contained continuant consonants that could be stretched and connected. Following learning to criterion, students completed a transfer task to decode CVCs with stop consonants that are harder to blend because of intrusion from schwa vowels. Results showed that connected phonation training facilitated learning to decode as well as reading nonwords accurately on the transfer task compared to segmented phonation training. An error analysis suggested that breaking between phonemes caused students to forget initial phonemes during blending. Findings suggest how to teach decoding more effectively.