You usually need to run a few different assemblers and see what works best with your data. If you have 2x150bp reads from a single PCR-free library based on gel-free fragment selection, you could try DISCOVAR de novo. Although the DDN authors recommend 250 base reads, reads as short as 150 bases may work. Other options might be SPAdes, SGA, ABySS 2, Meraculous2, and MaSuRCA.
In my experience you can get medium-sized insect genome assemblies with good gene content and contiguity by correcting reads with BFC, assembling contigs with SPAdes (turning off its error correction module, BayesHammer), scaffolding with SGA, and fixing errors with Pilon. If your insect genome is actually much larger than 300Mbp, using SPAdes is probably not a good idea. Platanus is another option, specially if the genome is highly heterozygous, although in my experience you get very poor results with a single paired-end library; you would need reads from at least one mate-pair library. ALLPATHS-LG is another alternative if the paired-end reads overlap and you have at least one mate-pair library. If perhaps you can sequence long reads, you could try a hybrid assembly with SPAdes or other assemblers, too.