One of the most well-known instructional methods used to enhance Chinese character reading (CCR) is Chinese stem-deriving instruction (CSDI). In this method, CCR is taught via a group of characters (e.g., [foreign character omitted], [foreign character omitted], and [foreign character omitted]) that share the same stem (e.g., [foreign character omitted]). However, the effectiveness of CSDI on the CCR of Chinese children with dyslexia remains inconsistent, and the reason may be the diverse designs of this method across studies. Thus, this study aimed to use a quasiexperiment-design study to compare the effects of various forms of CSDI based on the (in)consistency of the extending characters' sounds, their appearances, and their amount (i.e., how many of them). In total, 122 Chinese third to sixth graders with dyslexia were recruited from primary schools, and they were randomly assigned to one of three groups: (a) visuospatially controlled CSDI (n = 37; all taught characters shared the same pronunciation), (b) verbally controlled CSDI (n = 45; all taught characters shared the same structure), and (c) noncontrolled CSDI (n = 40; all taught characters selected only on the basis of their frequencies). They were tested by CCR and dictation before and after the instruction. We found that CCR was significantly improved in all three CSDI groups, whereas only the verbally controlled CSDI group showed significant improvements in dictation. Furthermore, verbally and visuospatially controlled CSDI earned the greatest benefits in decreasing the visual errors and phonetic errors in CCR and decreasing the radical errors and homophone errors in dictation. Our findings shed light on the effectiveness of CSDI in terms of diverse designs, and frontline practitioners should take our findings into consideration when using CSDI.