You just finished a Matplotlib session. You can call bar() and hist() in your sleep. Now you're staring at a dataset, wondering which one actually belongs there.
Both charts use rectangles. Both sit on an x-axis and y-axis. Both can be teal if you want them to be. Picking the wrong one doesn't just look sloppy — it actively misleads whoever reads your chart. That's the real stakes here, not syntax.
The one-question test that decides your chart
Forget the code for a second and ask yourself one question: are you comparing separate groups, or are you looking at the spread of one continuous variable?
If someone could reasonably ask "which one is bigger?" about your data — categories, regions, product lines, months — reach for a bar chart. If someone is asking "what's the range of values here, and where do most of them cluster?", you're dealing with a distribution question, and that's histogram territory.
The giveaway is usually the x-axis. Bar charts put discrete labels on the x-axis: North, South, East, West. Histograms put numeric ranges on the x-axis: 20-30, 30-40, 40-50. If you can shuffle the order of your x-axis categories without losing meaning, you're in bar chart land. If shuffling would break the story — because 20-30 has to come before 30-40 — you're dealing with continuous data and a histogram.
There's also a visual tell you can spot from across the room. Histogram bars touch each other by convention, because the underlying variable is continuous and there's no real gap between one age range and the next. Bar charts leave visible space between bars, because each one represents a separate, unrelated category. If you see gaps, someone's comparing categories. If the bars sit shoulder to shoulder, someone's showing a distribution. See more on how to read open-source code like a senior engineer for additional background.
Same data, two charts, two very different stories
Here's where it gets concrete. Take a list of 20 customer ages, the same ones from a typical intro session: 22, 25, 25, 28, 30, 31, 31, 31, 35, 36, 38, 40, 42, 45, 45, 48, 50, 52, 55, 60. Rendered as a histogram, you're asking Matplotlib to group these into ranges and show you the shape of the distribution.
import matplotlib.pyplot as plt
ages = [22, 25, 25, 28, 30, 31, 31, 31, 35, 36, 38, 40, 42, 45, 45, 48, 50, 52, 55, 60]
fig, ax = plt.subplots(figsize=(7, 5))
ax.hist(ages, bins=6, color="steelblue", edgecolor="black")
ax.set_title("Customer Age Distribution")
ax.set_xlabel("Age")
ax.set_ylabel("Number of Customers")
plt.show()
That chart tells you the ages cluster in the low 30s and thin out toward 60. It's a shape story: skewed, clustered, spread out, whatever it happens to be.
Now take that same list and sort it into categories — say, marketing-style age bands: 18-29, 30-39, 40-49, 50+. Count how many customers fall in each band, and now you've got a bar chart.
band_labels = ["18-29", "30-39", "40-49", "50+"]
band_counts = [3, 6, 6, 5]
fig, ax = plt.subplots(figsize=(7, 5))
ax.bar(band_labels, band_counts, color="teal")
ax.set_title("Customers by Age Band")
ax.set_xlabel("Age Band")
ax.set_ylabel("Number of Customers")
plt.show()
Both charts come from the same 20 numbers, but they answer different questions. The bar chart is punchy and easy to read in a slide deck: "30-39 and 40-49 are our biggest customer bands."
The histogram is more honest about the underlying shape. It shows a real cluster around 30-31 that the band chart smooths over completely. If you only saw the bar chart, you'd never know that three of the six customers in the "30-39" band are all 31 years old.
That's not a hypothetical concern. Grouping continuous data into categories for a bar chart is a design decision, and it always throws away information. Sometimes that's fine, even useful, because stakeholders don't need to see every data point's exact position. Other times it hides the very pattern you're trying to reveal.
Bins are opinions, and they rewrite the narrative
Even when you commit to a histogram, you're not done making judgment calls. The bins parameter isn't a technical afterthought — it's an editorial decision.
Set bins=3 on that same age list, and Matplotlib collapses everything into three wide buckets, smoothing out any bumps and making the data look more uniform than it is. Set bins=15, and you get jagged, noisy bars, because you're asking for more precision than 20 data points can support. Somewhere in between — bins=6 in this case — you get a shape detailed enough to be informative but not so fragmented that it's just noise.
Also read: claude on aws: bedrock vs platform security checklist in depth
There's no universal right number of bins. A commonly cited rule of thumb is the square root of your sample size, which for 20 points lands around 4 or 5, but treat that as a starting point, not a rule. The real test is whether the shape you're seeing would survive a small change in bin count. If your "skewed distribution" disappears when you bump bins from 6 to 8, you weren't looking at a real pattern — you were looking at bin noise.
The same scrutiny applies to bar charts, just in a different form. If you're grouping continuous data into bands to build a bar chart, the boundaries you pick — 18-29 versus 18-30, 40-49 versus 35-44 — can shift which category looks biggest. That's not dishonest by default, but it's worth being deliberate about, especially if you're building a report someone else will act on.
Once you internalize the real question — categorical comparison or continuous distribution — the choice between bar() and hist() stops being a coin flip and starts being obvious almost every time. The harder, more valuable skill is the one nobody mentions in the syntax docs: deciding how to bin your data, and being honest with yourself about what that decision hides. Get that right, and the chart type basically picks itself.
