If we calculated two means, graphed them and found that the means differ numerically, how could we know whether their difference is significant? This is especially important when comparing a control to an experimental group. Standard error bars can be added on both sides. In general, the greater the degree of error bar overlap the less a chance of a significant result. A non-significance result would mean that statistically there is not difference between the two means. If there is an overlap between the error bar of one mean and the mean of the compared mean, then it is not likely the means are significantly different. If error bars slightly overlap, significance might have to be additionally verified by a t-test. If error bars do not overlap at all, the results may be considered significant. Also, as sample size decreases the more error should be reported. So as a general rule if the sample size is large then +/- 1 SEM bars would be added. If your sample size is small (<30 samples), using 95% CI (~2xSEM) is preferred.