Translation of messenger RNA (mRNA) into protein is a critical process to cell survival and function. mRNA translation is a tightly regulated process, most of which occurs during the initiation stage. Eukaryotic translation initiation primarily relies on a cap-dependent scanning mechanism for start codon selection, facilitated by over a dozen of eukaryotic initiation factors (eIFs) and several RNA helicases. The current model of eukaryotic translation initiation commences with the attachment of pre-initiation complex (PIC) to the 5’ end, followed by 5’_3’ uni-directional scanning. Despite the proposed scanning process in start codon selection, the dynamic nature of the scanning PIC is poorly understood. Investigation of start codon recognition using synthetic mRNA reporters with ultra-short 5’UTR or closely spaced AUG triplets suggests that the PIC follows a non-linear scanning mode. I demonstrate the backward excursion of the scanning ribosome using an internal ribosome entry site that enables translation of both upstream and downstream open reading frames. Unexpectedly, the ATP-dependent DEAD-box RNA helicase eIF4A is actively involved in bi-directional oscillations during scanning. Using quantitative profiling of initiating ribosomes (QTI-seq), I show that upregulation of eIF4A ATPase activity globally regulates alternative initiation by stimulating translation from the upstream initiation sites in the 5’ untranslated region (5’UTR). Moreover, this study revealed that eIF4A1 mediated bi-directional ribosome scanning controls the stringency of start codon selection. During the scanning, the PIC has to migrate along the structured region in 5’UTR, with the help of various RNA helicases. The DEAD-box helicase eIF4A1 was reported to remodel the 5’ proximal structures with low stabilities. Another RNA helicase, DHX29, was known to promote unwinding of more stable secondary structures. However, their potential roles in translation of mRNAs with secondary structures in different regions remain elusive. Herein, I reported that the position of stem-loops (SLs) in the 5’ UTR did not affect the dependence of mRNA translation on either eIF4A1 or DHX29. Surprisingly, the translation of mRNAs with SLs downstream of the start codon were impaired by DHX29 depletion. DHX29 was found in both 40S and 80S ribosome fractions after polysome fractionation, supporting the presence of DHX29 in 80S. Collectively, these data suggest a novel function for DHX29 in unwinding stem loops during the early stages of elongation. In summary, this study reported the bi-directional motion of PIC during scanning, which potentially could address the questions regarding the mRNA attachment and the start codon stringency. The novel functions of eIF4A1 in the scanning directionality via ribosome conformation and DHX29 in unwinding mRNA secondary structures during early elongation were demonstrated. This body of work might contribute to better appreciation of the wide-scope functions RNA helicases may have, as well as the comprehensive understanding of the PIC scanning during initiation.